Tracing the Distributional Effect of Trade Policy on Firm Value

Tracing the Distributional Eﬀect of Trade Policy on Firm Value∗

Yahui Wang† This Version: December 2019

Abstract

I investigate the effect of trade liberalization policy on firm value. I identify this effect by exploiting cross-sectional differences in firms’ exposure to potential tariff hikes imposed on U.S. imported goods from China. I examine both U.S. and Chinese equity market responses when U.S. granted permanent normal trade relations (PNTR) to China in 2000, and document puzzling responses from the Chinese market: Firms with larger downside risks that were eliminated—and thus were likely to gain more access to the U.S. market—suffered larger equity value reduction when the PNTR bill passed. I discover that the negative stock market reactions are driven mainly by state-owned enterprises (SOEs). I then propose a mechanism in which SOEs and non-SOEs differ in their capacity to withstand negative shocks due to government’s safety net provision. The proposed mechanism is supported by evidence from firm exporting activities and stock market responses during anti-dumping episodes. My results imply that trade policy uncertainty elimination may generate distributional gains from reallocation to more efficient firms in the context of inefficient institutions.

∗I am indebted to my advisor, Kent Daniel, for his advice and support throughout this project. I am grateful to Jesse Schreger and Olivier Darmouni for their generous encouragement and instructions. I thank the participants in the Finance Seminar and PhD Student Seminar at Columbia Business School and the Financial Economics Colloquium at Columbia University. Any errors are my own. †Columbia Business School. Email: [email protected].

1 1 Introduction

The past two decades have witnessed China’s integration into the global trade market, which has been one of the most influential economic developments that shape today’s economic landscape around the world. China’s global power has grown considerably in recent years. It is now the world’s second largest economy by nominal GDP and the largest manufacturing economy and exporter of goods. For instance, the U.S.’s share of imports from China rose from 7% to 22% between 1997 and 2017.1 This paper investigates a major trade liberalization event—the U.S. granting permanent normal trade relations (PNTR) to China, which has drastically changed U.S.-China trade relations since 2000—and sheds light on how trade policy affects firms’ performance. I use an event study approach to capture changes in firms’ value from a capital market perspective and examine firms’ exporting activities to uncover the mechanism that explains their equity market responses to the policy shock. I confirm the results of Greenland et al.(2018), that U.S. firms experience greater market value reduction if they are more exposed to potential import competition from China.2 Surprisingly, I find that Chinese firms exhibit similar equity market responses in the cross- section: Chinese firms potentially gaining more access to the U.S. market seemed to suffer larger market value loss when the PNTR bill was passed. How could this U.S.-China trade liberalization event be a worse shock for Chinese firms supposedly gaining more market share in the U.S.? A closer look at the Chinese market reveals that the results of larger market value reduction with respect to higher degrees of U.S. market penetration are driven mainly by Chinese state-owned enterprises (SOEs). I propose a mechanism that features market share redistribution in response to foreign trade policy shocks to explain the puzzling Chinese equity market responses. SOEs enjoy safety net provisions from the government and were less sensitive to potential trade barrier increases before PNTR, and thus benefit less from the reduced trade cost induced by this trade liberalization. On the other hand, PNTR grants the entry of more Chinese non-SOEs into U.S. markets, driving down the profits of SOEs. The competition effect overshadows the benefit from reduced trade costs for Chinese SOEs, and contributes to the negative equity market responses. This is the first paper, to the best of my knowledge, that documents Chinese firms’ equity

1Source: United Nations Statistics Division (UNSD) 2I follow Pierce and Schott(2016) and quantify potential increases in import competition after the PNTR event via the “NTR gap,” defined as the difference between non-NTR tariff rates and NTR tariff rates.

2 market performance around the PNTR event. The U.S. granting PNTR to China in October 2000 is a much-studied policy change in the international trade literature, yet few papers explore the capital market consequences of this U.S.-China trade liberalization. This paper fills the gap. It also provides first evidence on how the Chinese equity market reacts to the PNTR shock, and further rationalizes the findings in the context of insufficient institutions. I adopt an event study approach and look at stock market reactions around the PNTR event. The benefit of using equity market performance is that it captures the expected net influence on firms (and industries). Trade policy changes not only affect current economic outcomes, but also shape investors’ expectations on future cash flows and discount rates, which are reflected in the stock prices. In an informational efficient market, the stock price movements reflect the market implied overall impact of the trade policy shock on firms’ cost of capital. The event study approach, on the other hand, is justified because there was considerable doubt regarding passage of the China Trade bill granting permanent NTR status to China. The legislation received Republican support, but the Democrats voted against it.3 It took the combination of a Republican Congress and a Democratic president to successfully pass the bill.4 The conferral of PNTR in 2000 eliminates the threat of tariff increases being imposed on U.S. imports from China. Before this policy shock, normal trade relations (NTR)5 status for China was temporarily granted and required annual renewal by the U.S. Congress, beginning in 1980. Had China’s NTR status been revoked, U.S. import tariffs on Chinese goods would have jumped to a much higher set of non-NTR tariff rates6, which were established under the Smoot-Hawley Tariff Act of 1930. I define “NTR gap” as the difference between the higher non-NTR and the lower NTR tariff rate, following Pierce and Schott(2016), for all U.S.-China bilaterally traded goods for which both sets of tariff rates are available. I then aggregate the product-level tariff spreads to compute firm-level NTR gaps to quantify the degree of potential import competition for U.S. firms (and, equivalently, U.S. market access

3Detailed voting records can be accessed at https://www.govtrack.us/congress/votes/106-2000/ h228. 4In the last year of his presidency, Bill Clinton called on Congress to help improve trade relations with China. 5Normal trade relations (NTR) is a U.S. term for the familiar principle of most favored nation (MFN). MFN is a status or level of treatment accorded by one state to another in international trade. The members of the World Trade Organization (WTO) agree to accord MFN status to each other. Since 1998, the term “normal trade relations” (NTR) has replaced most favored nation (MFN) in all U.S. statutes. 6In 2000, the average MFN tariﬀ rate was less than 5%, while the average non-NTR tariﬀ rate was around 30%.

3 for Chinese firms). Within the framework of the event study, I investigate firms’ equity returns across 2 days before and after five key legislative milestones.7 I find that, over the 25 days within the voting event windows, a one standard deviation increase, 14%, in the NTR gap corresponds to approximately a 5.2% (2.4%) decrease in market value, or about 160 million USD (93 million CNY, or 11 million USD8) loss for shareholders in the U.S. (Chinese) equity market. After documenting firms’ stock returns in response to the PNTR policy shock for both U.S. and Chinese markets, I further explore how abnormal returns could carry negative loading on the scale of freed-up U.S. market access for Chinese firms. I find that the surprising negative responses are mainly driven by Chinese SOEs. The PNTR bill has distributional effects, and hurts certain types of Chinese firms. Namely, for Chinese firms operating in an industry (A) that has one standard deviation more exposure to the U.S. market than another industry (B), the magnitude of underperformance by SOEs versus non-SOEs is approximately 6% more—roughly 216 million CNY (or 26 million USD)—in industry A than in industry B. I then propose a mechanism to rationalize the distributional effects observed in the Chi- nese market. I model firms’ differential sensitivity to negative policy shocks within a simpli- fied Melitz(2003) framework of heterogeneous firms. Simulations demonstrate that Chinese firms’ profits decrease in larger NTR-tariff-gap industries, but only for SOEs, which is consistent with the equity market evidence. The model also predicts contrasting exporting behaviors for Chinese SOEs and non-SOEs. Using administrative-level Chinese customs data, I provide empirical support for the proposed mechanism. Consistent with the model, I find faster growth in the number of exporters and a larger increase in the value of exported goods in larger NTR-gap industries, but only for non-SOEs. SOEs, on the other hand, cut down exporting activity on both the intensive and extensive margin in larger NTR-gap industries. It is worth pointing out that the majority of the empirical results explore the NTR gap distribution in the cross-section. The fact that about 80% of NTR gap variation comes from non-NTR tariff rates, which were established under the Smoot-Hawley Act in 1930, renders it more likely that NTR gaps are exogenous to the economic conditions in the 21st century. First, the equity market event study could be deemed a difference-in-difference design in disguise, in which I measure firms’ abnormal return relative to a benchmark (first difference)

7The choice of event window follows Greenland et al.(2018). 8The exchange rate, in force between January 1997 and July 2005, is 8.28 CNY per USD.

4 in the cross-section of different levels of NTR gaps (second difference). Moreover, my analysis of Chinese firms’ exporting activity before and after the China Trade bill (first difference) also employs a difference-in-difference setting that compares firms with different NTR gaps (second difference).9 Since the key mechanism that explains the distributional effect on Chinese firms’ equity market performance is through different capacities to absorb negative shocks for distinctive types of firms, I empirically justify SOEs’ partial absorption of potential negative shocks10 by examining their stock returns around realized negative trade shocks. The specific realized negative trade shocks I study are anti-dumping investigations of Chinese imports into the U.S. market. Using an event study approach, I find that Chinese SOEs’ stock returns are less affected than non-SOEs when they are exposed to anti-dumping investigations. This provides direct evidence that SOEs are less affected by bad trade shocks, possibly through safety net provision by the government.11 I discuss the implications and argue that trade liberalization can trigger additional gains by encouraging the entry of more efficient firms, which is especially salient in the context of developing countries with inefficient institutions. On the other hand, this research may provide a valuable reference in discussions of how today’s trade protectionism—under the Trump administration, for example—could generate additional losses due to less efficient allocation of resources, especially for developing economies.

Related Literature This paper relates to the expanding literature on the China trade shock. Autor, Dorn and Hanson(2013) document the “China syndrome” of the U.S. facing import competition due to the China trade shock. Auer and Fischer(2010) ﬁnd that the share of U.S. manufacturing expenditure on Chinese goods has contributed to declines in U.S. prices. Autor, Dorn and Hanson(2016) show that U.S. manufacturing employment and local wages also suﬀer from the rapid increase in Chinese import into the U.S.. I study the origin—in other words, the U.S. legislation granting PNTR, which paved the way for China’s accession to the WTO and

9In examining product-level Chinese exports, I use a triple-difference design in which the third dimension of variation comes from whether the exporting destination is the U.S. or not. 10Before the China Trade bill in 2000, the U.S. did not actually change the import tariff rates applied to Chinese goods. Therefore, Chinese firms essentially faced a potential threat of negative trade shocks, i.e., a sudden tariff increase to the Smoot-Hawley non-NTR level. 11Anecdotal evidence also suggests that the Chinese government provides special treatment or trade subsidies to SOEs.

5 further integration into the international trade market. In contrast to the large literature on the labor market effects of the trade liberalization with China (Auer and Fischer(2010); Autor et al.(2014); Acemoglu et al.(2016); and Pierce and Schott(2016), to name a few), few papers exploit the capital market consequences of the U.S.-China trade policy. Instead of trade liberalization, Huang et al.(2019) studies firms’ stock market responses to the current U.S.-China trade war, which is a form of protectionism. My paper, in contrast, investigates capital market reactions to a U.S.-China liberalization policy. Recently, The stock market response to the PNTR shock in particular has drawn atten- tion. Esposito, Bianconi and Sammon(2019) argue that the U.S. granting PNTR to China resolves trade uncertainty, and find that U.S. manufacturing industries that are more exposed to the uncertainty have higher stock returns. Griffin(2018) also explores the equity market response to PNTR, and suggests that increased trade with China led to a decline in U.S. public listing rates and increase in industry concentration. Greenland et al.(2018) use firms’ average abnormal return to infer their exposure to trade liberalization. My paper differs in that it compares both U.S. and Chinese firms’ equity market response to the PNTR event. This is the first paper to document Chinese firms’ stock reaction to the historical U.S.-China trade liberalization shock. Abstracting from the specific case of US-China trade, inferring winners and losers from international trade is an interesting question in and of itself. Gains from trade might arise from multiple causes, such as specialization in factors of production and and increase in total output possibilities (Krugman(1979); Krugman(1980); Krugman(1991)). This paper examines the gains from trade from a capital market perspective. Using equity market performance to infer distributional gains from trade is a novel addition to the literature on measuring gains from trade. Abnormally high stock returns could provide a direct assessment of the net impact for firms, because they proxy for return to capital (Grossman and Levinsohn (1989)), while in most traditional trade liberalization research, scholars use changes in tariffs and import volumes as measure of import competition to identify the effect of trade policy shocks (Autor et al.(2014); Hakobyan and McLaren(2016)). My investigation of the PNTR also relates to the literature on how economic and policy uncertainty12 affects financial markets. On a macro level, Bansal et al.(2014) find that un-

12Conferral of PNTR essentially eliminated trade policy uncertainty between U.S. and China at that time, because China’s NTR status with the U.S. would not be subject to annual approval by the Congress once

6 certainty plays a significant role in explaining the joint dynamics of returns to human capital and equity. Micro-level firms delay investments because uncertainty creates an option value of waiting (Dixit(1989)). In the context of international trade, Handley and Limão(2015) and Handley and Limão(2017) document that policy uncertainty plays an important role in firms’ investment behavior and economic outcomes. While the literature on uncertainty is mainly focused on the effect of increased uncertainty (Bloom(2009)), the unique setting of the PNTR policy shock is a case in which policy uncertainty is reduced (or even removed). Esposito, Bianconi and Sammon(2019) also study the policy uncertainty associated with U.S. granting permanent NTR status to China and explore its impact on the U.S. stock market over a long period of time. My paper, on the other hand, takes an event study approach and combines observations from both the Chinese equity market and the Chinese exporting activities. My paper offers a new mechanism for how trade policy (or even policy uncertainty resolution) could potentially improve aggregate welfare in developing countries, where insufficient institutions often lead to market distortions. Such misallocation can generate large economic losses. Hsieh and Klenow(2009), for example, calculate that distortions account for to 30% - 50% lower aggregate total factor productivity in China. Khandelwal, Schott and Wei(2013) propose that trade liberalization induces additional gains from the elimination of the embedded institution in addition to the removal of the trade barrier itself. This paper provides similar evidence, but differs along several dimensions. First, I focus on the policy uncertainty elimination instead of outright tariff reduction liberalization. Second, Khandelwal, Schott and Wei(2013) study only the textile industry, while this paper considers a richer cross-section of industries exposed to the U.S.-China trade liberalization shock. Finally, this paper contributes to the literature on corporate safety nets and financial distress. Due to preferential treatment, different types of firms have asymmetric sensitivity to the (potential) negative shocks. Faccio, Masulis and McConnell(2006) document that politically connected firms are more likely to be bailed out than similar but unconnected firms. Politically connected firms are backed by a government safety net, and thus absorbing only a fraction of the impact if a bad shock hits. To the best of my knowledge, this is the first paper that explores the interaction between policy uncertainty shocks, international trade, and corporate default implications with favorable downside risk protection from inefficient

the permanent NTR bill was passed. Here the “uncertainty” terminology refers to the indeterminacy of the policy due to the annual renewal process by the U.S. Congress, and not to the commonly received notion of variance.

7 institutions. The paper proceeds as follows: Section2 and section3 describe the institutional background and data. Section4 presents the empirical analysis and main results, with a focus on surprising ﬁndings in the Chinese equity market. These are rationalized by the mechanism proposed in Section5, which features a qualitative trade model to illustrate the mechanism. Section6 contains more empirical results consistent with the theoretical prediction, and provides empirical evidence to justify the mechanism at work. Section7 discuss the implications and then conclude.

2 Policy Background: PNTR

U.S. tariff schedules currently have two major sets of rates: a relatively low set of “column 1” rates, which are offered to NTR13 partners, and a higher “column 2” set of rates for non- NTR partners. U.S. imports from non-market economies, for example, are charged the higher non-NTR tariff rates originally set under the Smoot-Hawley Tariff Act of 1930, which was signed by President Hoover to fulfill the trade protectionism promises he made during his presidential election campaign. The U.S. government later reduced the tariff rates on imports from most trading partners, but non-market economies are still subject to the higher tariff unless a waiver is approved by Congress. Such a waiver is usually effective only temporarily, and requires annual renewal. China first received the temporay NTR tariff waiver in 198014 and enjoyed the relatively low tariff rate as a result. The waiver was successfully renewed without issue for almost a decade until the Tiananmen Square event in 1989. The U.S. government began to reevaluate its trade relations with China, and considered revoking China’s temporary MFN status. Although the revocation did not happen, anecdotal evidence indicates that the threat to end the NTR waiver was taken seriously (Pierce and Schott(2016)). The policy uncertainty regarding whether China would lose its NTR status continued for the next decade15, and was suddenly resolved by the China Trade bill in 2000 granting permanent NTR status to China. 13In the U.S., normal trade relations (NTR) refers to most favored nation (MFN) before the change of terminology in 1998. 14Because China was viewed as a non-market economy, its MFN/NTR status was originally suspended in 1951, restored in 1980, and continued in effect through subsequent annual extensions. 15The average House vote against the NTR renewal was around 40%.

8 Five legislative milestones led to the passage of the PNTR bill: (1) introduction of the bill in the U.S. House of Representatives (2) passage by the House (3) motion of cloture in the Senate (4) passage by the Senate (5) signing of the bill by President Clinton. I use all of these five legislative event dates in the study. In addition to tariff-based liberalizations, the PNTR agreement also requires Chinese nontariff commitments such as the phase-out of trade quotas and licensing requirements. Following China’s accession to the WTO, prevalent anti-competitive behavior by SOEs becomes an actionable offense.

3 Data

3.1 International Trade: HS code and Concordance

The Harmonized System (HS) of tariff nomenclature, also known as the Harmonized Com- modity Description and Coding System, is widely used by customs authorities and various government regulatory bodies to classify and monitor international trade of commodities. The HS is an internationally standardized system of names and numbers to classify trade products. The HS is organized by economic activity or component material into 21 sections, which are subdivided into 99 chapters and further into over 1,000 headings and over 5,000 subheadings. HS chapters, headings, and subheadings are subsequently represented by 2- digit numbers. As a result, the basic HS code consists of six numerical digits. The 6-digit HS codes can be locally extended to eight or 10 digits for further tariff discriminations in some countries and customs unions. Feenstra, Romalis and Schott(2002) provide U.S. tariff rates data at the HS level. 16 One can compute the difference between the ad valorem equivalent17 MFN rate and the non-MFN rate of the Harmonized Tariff Schedule given a specific HS code. Pierce and Schott(2016) did this calculation and coined the tariff difference as the NTR gap. Using the HS products’ NTR gaps, I construct the implied NTR gap for each NAICS industry.18 I map product-level HS codes to their corresponding NAICS industries using

16NBER trade data are available at www.nber.org/data/. 17When a tariff is not a percentage (e.g. dollars per ton), it can be estimated as a percentage of the price. 18The North American Industry Classification System (NAICS) is the standard used by federal statistical agencies in classifying business establishments for the purpose of collecting, analyzing, and publishing statistical data related to the U.S. business economy. The 2017 version NAICS manual can be downloaded from https://www.census.gov/eos/www/naics/. The North American Industry Classification System (NAICS) officially replaced the Standard Industrial Classification (SIC) code system in 1997. While SIC codes are

9 the concordance provided by the U.S. Bureau of Economic Analysis (BEA) and Pierce and Schott(2009). Next, I compute simple and import value weighted averages of the NTR gaps across all HS products within the same NAICS code and derive the NAICS-level NTR gaps. HS product-level imports from China to the U.S. are downloaded from the United Nations International Trade Statistics Database (UN Comtrade).

3.2 Equity Markets: U.S. and Chinese Firms

3.2.1 U.S. Firms

I start with the universe of publicly listed U.S. firms in the CRSP database that can be matched to Compustat. I filter for ordinary common shares (share code equals 10 or 11) listed on the NYSE, AMEX, and Nasdaq. Historical NAICS and firm-level accounting variables are obtained from Compustat. Product-level information for each firm is obtained from Compustat Segment Data, which covers over 70% of the companies in the North American database. I drop firms whose trading data are not sufficient19 for the pre-period market beta estimation. The final combined sample contains 5,160 firms, of which 2,386 firms have nonmissing self-reported sales data on product segments. Firms with missing return data on any of the five event dates can either be dropped or have their missing window imputed from the average return calculated using the rest of the windows where return data is available. The main empirical findings are similar to those specifications and are discussed in the robustness section. Daily U.S. market excess returns are retrieved from Kenneth French’s web page.20

3.2.2 Chinese Firms

I combine the following datasets to analyze the Chinese market. Chinese stock data come from two major sources. First, Wind Information Inc.(WIND), known as the Bloomberg of China, serves approximately 90% of China’s ﬁnancial institutions in China and has comprehensive data on returns and trading. Second, CSMAR (China still used by some organizations and government agencies for nonstatistical purposes, NAICS numbers are the standard for federal economic study applications. 19Firms need to be present for at least 120 days of the 250 trading days in 1999, the year previous to passage of the PNTR bill. Alternatively, using the 250 trading days that end one month before each legislative event does not change the main results qualitatively or quantitatively. 20The data library can be found at https://mba.tuck.dartmouth.edu/pages/faculty/ken.french/ data_library.html.

10 Stock Market & Accounting Research) is a database for Chinese business research, with data on both the Chinese stock market performance and financial statements of Chinese listed firms. I use CSMAR as the main data source for Chinese firms’ stock returns and financial statements, while cross-checking the sample with those obtained from WIND. My sample includes all A-share and B-share21 stocks from the main boards of the Shanghai and Shenzhen exchanges. Stock tickers issued by the China Securities Regulatory Commission (CSRC) are non-reusable unique identifiers for listed firms in China. The first two digits of the 6-digit stock ticker indicate the exchange (Shanghai or Shenzhen) and the security type (A-share or B-share). I then apply the same filter as I process the U.S. stock market data to exclude stocks having fewer than 120 days of trading records during the market beta estimation period. My Chinese firms sample reduces to 899 public firms, with 981 distinct stock identifiers due to multiple listings on both stock exchanges and possibly different security types. I also drop firms whose trading data are either missing for all of the legislative event dates or are inadequate for the pre-period market beta estimation. The second source is the annual surveys of above-scale industrial firms collected by the National Bureau of Statistics (NBS) of China. The Chinese Industrial Survey, which is similar to the U.S. Census of Manufacturing, contains plant-level micro data on manufacturing establishments in China. The survey covers all SOEs and non-SOEs with sales above 5 million CNY (or 604,000 USD).22 I obtain each firm’s industry classification from this annual survey.23 I combine the above datasets by matching firms’ Chinese names. Firms sometimes share or change their name; as a result, I did a manual check of Chinese firm names to ensure reliable matches. For example, when multiple firms share the same name in the industrial census but only one firm is listed on the stock market, I manually check other firm information listed in the manufacturing census—firm address, ownership, etc. Other cases also require manual matching. For example, almost one-third of the companies experienced drastic name changes due to mergers and acquisitions, changes of primary business type, or reverse takeover ac-

21A shares are ordinary shares freely traded in Chinese yuan (CNY) on Chinese stock markets, and are available for purchase and sale by the mainland residents. B shares are foreign investment shares traded on Chinese stock markets. The face value of such shares is in CNY, but trading is conducted in foreign currencies. These are shares for trading by investors from Hong Kong, Macao, Taiwan, and overseas. 22The raw data consist of over 100,000+ plants in year 2000. 23 Though the observations are recorded at the plant level, more than 95% of all observations in the annual survey data are single-plant ﬁrms.

11 tivities. Finally, reading though companies’ annual reports and announcements enables me to build a reliably matched sample, which consists of 779 companies. Daily Chinese market excess returns are retrieved from the CSMAR Trading Database.

3.3 Industry Classiﬁcation

The U.S. firms in my sample are classified into various industries according to NAICS. While the earlier SIC system used a market-oriented logic, the NAICS system employs a production-oriented structure that identifies businesses only by their primary economic activity. When a firm’s product segment data is available through COMPUSTAT, it may contain multiple NAICS assignments for each of its product segments. Hence, the NAICS system is a better choice than the SIC for this study. For Chinese firms, the official China Industry Classification (CIC) maintained by China’s National Bureau of Statistics (NBS) is the most commonly used and serves as the national standard. The Industrial Classification for National Economic Activities is the GuoBiao (GB) Standards24, which lists the CIC from China’s NBS. There are four digits, corresponding to section, division, group, and class. The CIC classification went through several major revisions to keep up with the changing economic activities most relevant to the Chinese economy. The GB Standard was launched in 1984, and underwent its first major revision in 1994. From 1995 to 2003, companies included in the NBS manufacturing census were encoded with the GB/T 4754-1994 standard. From 2003 to the third revision in 2011, CIC codes were assigned according to the GB/T 4754-2003 standard. I combine the official conversion guide and concordance by Brandt, Van Biesebroeck and Zhang(2014). 25 For publicly listed companies on the Chinese market, there are alternative sources of industry classification. The industrial codes used in the CSMAR dataset are assigned by the China Securities Regulatory Commission (CSRC). The CSRC uses the Guidelines on Industry Classification of Listed Companies26 to classify the economic activities of the listed companies based on sources of their operation income. The CSRC’s classification guidelines

24GB standards are China’s national standards, also known as Guobiao Standards. The latest GB Stan- dards (GB/T 4754-2017) on the Industrial Classification for National Economic Activities can be purchased and downloaded at http://www.gbstandards.org/GB_standards/GB-T%204754-2017.html. 25The official conversions between the 1994 and 2000 revision can be accessed at http://www.stats. gov.cn/english/18round/papers/200501/t20050114_44579.html. The harmonized classification constructed by Brandt, Van Biesebroeck and Zhang(2014) can be found at https://feb.kuleuven.be/public/ u0044468//CHINA/appendix/. 26The Guidelines are available at http://www.csrc.gov.cn/pub/csrc_en/newsfacts/release/200708/ t20070816_69104.html.

12 are derived from the CIC system. However, there are several disadvantages to using this classification directly. First, the CSRC Guidelines have coarser categories compared with the CIC from China’s NBS. Second, the CSRC classification obtained through CSMAR is not historical, and thus not immune to changes in business operation over the years. For example, a shoe manufacturing firm in 2000 that was restructured into a metal production company in 2015 would be classified in the metal production industry when we access the data in 2018. Therefore, the industrial classification innate in the CSMAR database should not be used directly for analysis. In order to make consistent comparisons between U.S and Chinese firms operating in the same industry, I map the NAICS to China’s CIC using concordances created by Ma, Tang and Zhang(2014), which contain 525 unique 4-digit CIC and 511 unique 6-digit NAICS categories. Constructing the Chinese NAICS-equivalent industrial NTR gaps entails the following process27. First, I compute the NAICS-level NTR gaps in the same manner as for the U.S. firms. Next, I compute for each CIC assigned to Chinese firms the CIC-level NTR gap based on the above concordance. When one NAICS industry corresponds to multiple CIC industries, adjustments are made by equally reducing the weight of the NAICS industry among all matching CIC industries. This allows me to assign NAICS-equivalent NTR gaps to 210 unique 4-digit CIC Chinese industries.

3.4 Chinese Exporting Activities

I obtain highly disaggregated product-level trade data from Chinese customs, covering 2000 to 2012. Each firm-product-country entry in the database provides detailed trade information, including products’ export value, shipment method, exporting destination, etc. In the Chinese customs data, exporting firms are also classified by their administrative types into SOEs and non-SOEs. 27An alternative method entails the following four steps. First, NTR gaps are computed at the 8-digit HS level provided by Feenstra, Romalis and Schott(2002). Second, I map product-level HS codes to their corresponding CIC industries using the concordance obtained from Brandt et al.(2017). The concordance is between the trade classification for productions (at the 6-digit HS level) and the industry classification used in the NBS firm-level data (at the 4-digit CIC level). Next, I take the simple average and two versions of value-weighted mean of the NTR gaps across all HS products within the same CIC code and derive the CIC- level NTR gaps. The weighted average of the NTR gaps uses either the sum of imports’ general value or the imports value from China at each HS product level. Finally, I concord CIC-level NTR gap to NAICS-level NTR gap. The major results are robust to this specification.

13 3.5 Anti-dumping Investigations

The Temporary Trade Barriers Database (TTBD) includes detailed data on national governments’ use of policies such as anti-dumping (AD) and countervailing duties (CVD). The data, which I access from the World Bank, contain information on the global use of various import-restricting trade remedy instruments such as AD and CVD, including the ﬁling of U.S. investigative cases against Chinese imports.

4 Empirical Evidence

4.1 Average Abnormal Returns

Among the approaches to computing firms’ responses in an event study, a common method is to use the capital asset pricing model (CAPM) as the “benchmark28,” and a firm’s deviation of realized excess return from CAPM-implied expected excess return as the abnormal return. A firm’s excess return, under the CAPM benchmark, has two parts: the CAPM-implied expected excess return and the firm’s abnormal return. I use return data from all trading days in the previous year—i.e. 1999—before all of the event windows to estimate each firm’s CAPM β, as shown in Equation (1):29

E (Ri) = Rf + βi (E (Rm) − Rf ) (1) ˆ The abnormal return of firm i on date t, ARit, equals Rit − Rft − αî + βi (Rmt − Rft) , and averaging across the five30 days within one event window yields the average abnormal return for firm i, AARevent, equals AARevent = 1 Ptt+2 AR . The overall “PNTR” average i i 5 t=te−2 it PNTR abnormal return of firm i is then AARi :

tt+2 PNTR 1 X event 1 X X AARi = ARRi = ARit 5 event∈E 25 e∈E t=t −2 e (2) 1 tt+2 = X X R − R − αˆ + βˆ (R − R ) , 25 it ft i i mt ft event∈E t=te−2

28I also use market return as an alternative benchmark. The results are robust. 29Alternatively, using 250 days that end 30 days before each event does not qualitatively or quantitatively change the main results. Here I follow Greenland et al.(2018). 30Following Greenland et al.(2018), I choose a window two trading days before and after the ﬁve legislative event dates. Greenland et al.(2018) provide empirical support for the choice of the window length, based on the number of new articles in major news outlets that contain the word “PNTR.”

14 where the ﬁve events in E are House Introduction, House Vote, Cloture Motions, Senate Vote, and President Signing. Firms’ AAR distributions, both in the U.S and Chinese markets, can be found in the Appendices.

4.2 NTR Gaps in the Cross Section

U.S. imports from China were charged at the relatively low NTR tariff rates, which required annual renewal by Congress. The difference in tariff rates between the potentially higher non-NTR tariff rate, in case renewal was denied, and the lower NTR tariff rate is the NTR gap. The NTR gap arguably became exogenous to stock market performance around 2000, because approximately 80% of the variation comes from non-NTR rates, which were set in the 1930s. This unique feature of the NTR gap rules out the reverse causality concern that industries producing products with larger NTR gap would receive endogenous protection from higher non-NTR tariff rates. Feenstra, Romalis and Schott(2002) constructed the ad valorem equivalent NTR and non-NTR tariff rates set at the 8-digit HS level, also referred to as “tariff lines.” For each traded HS product h , one may calculate the NTR gap as

NT R Gaph = NonNT R Rateh − NT R Rateh (3)

Using concordances made available by the U.S. Bureau of Economic Analysis (BEA), I aggregate the product-level tariff rates to industry level by averaging the NTR gap across all 8-digit HS products within the same industry. I consider different weighting schemes when aggregating the product-level NTR gaps into the industry-level gaps. In addition to simple average by equal weights, one weighting method uses the sum of imports’ general value31 at each HS product level. Another weighting method focuses on the value of U.S. imports from China in particular. I use the NTR gaps for 1999 in the following analysis, because the PNTR shock occurred in 2000. However, the NTR gaps are relatively stable and not prone to change over the year. I also checked that the main results are robust to alternative choice of the years when NTR gaps are measured. Figures1 and2 show industry-level NTR gap distribution using different aggregation

31General imports measure the total physical arrival of merchandise from foreign countries, whether such merchandise enters consumption channels immediately or is entered into bonded warehouses or foreign trade zones under customs custody.

15 methods across the product tariff lines in my combined U.S sample and combined Chinese sample, respectively. Within the same industry classification, either under the NAICS or CIC, the density distribution looks very similar. The averages of NTR gaps across 6-digit NAICS industries are 0.30 for all specifications, with standard deviations in the range of 0.16 to 0.17. On the other hand, average NTR gaps across 4-digit CIC industries are between 0.27 and 0.29, with standard deviations from 0.15 to 0.16.

[Figure 1 about here.]

[Figure 2 about here.]

4.3 Regression Results: Equity Market Responses

I first examine the relationship between a firm’s average abnormal return and its exposure to the PNTR tariff shock. I regress firms’ AAR on firm-level NTR gap. For U.S. firms, when product sales segment data are available, a sales-weighted average NTR gap of all major segments’ 6-digit NAICS is used to compute the firm-level NTR gap. For diverse firms with product sales recorded in both good-producing and service industries, I assign a zero gap for the service segments before calculating the sales-weighted average. In the robustness section, I show that this product-level aggregation is not essential in driving the results. To simplify the computation, we may ignore the product segments’ weighting and use COMPUSTAT-assigned historical NAICS codes for the entire firm. For Chinese firms, on the other hand, reliable product segment sales data hardly exist. As a result, I use firms’ industry classification assigned by the NBS, which are included in the Manufacturing Survey data. The OLS specification is:

event AARi = α + βNT RGapi + γXi + i,

where Xi contains firm-level attributes controlling for investment and financing opportunities.32 32The control variables are firm size (as measured by the log of market capitalization), profitability (cash flows to assets), book leverage, and Tobin’s Q. This set of firm-level attributes follows Greenland et al. (2018).

16 4.3.1 U.S. Market Response

Table1 shows negative and statistically significant relationships between a firm’s average abnormal returns and its NTR gap, which proxy for the degree of potential import competition from Chinese firms. Figure3 visualizes the equity market responses of U.S. firms in the cross section. Here I present the main results by averaging across all five dates, while the study around each individual legislative event date is included in the robustness section. Combining all five events, I use the overall average abnormal return AARPNTR as the dependent variable, which is either CAPM adjusted, as in column 1, 3, and 5, or market adjusted as in column 2, 4, and 6 of Table1. There are three specifications for NTR exposure: equal weight, import-value weight, and import-value-from-China weight, which means that product-level NTR tariff spreads are either equally weighted or value weighted by total import value or import from China to compute the industry-level NTR gap.

[Table 1 about here.]

[Figure 3 about here.]

As indicated in Table1, the negative and statistically significant relationship between a firm’s average abnormal return and its NTR gap persists in all specifications. This negative relationship is also economically significant. Since all regression variables are standardized to have zero mean and unit standard deviation, the coefficient estimate in column 1 suggests that a one standard deviation increase in a firm’s NTR tariff gap exposure corresponds to a 0.186 standard deviation decrease in daily average abnormal return during the event windows. This translates to a total of 5.16% decline in market value, or about 160 million USD loss for shareholders.33 One may worry about the benchmark we used in computing the average abnormal returns during the event windows. In particular, firms with higher market beta during the pre-event estimation window might obscure the regression results. Stocks might have low CAPM- adjusted abnormal returns simply because they have high market betas. Even though it is hard to explain why high-beta stocks behave differentially according to different PNTR tariff

33From the regression estimate with standardized variables, we can multiple the point estimate of -0.186 PNTR by the standard deviation of AARCAP M , 1.11%, then multiply by 25 (number days of the event window), resulting a total reduction of 5.16%. Considering that the average market value of a ﬁrm in 2000 in my sample is 3.1 billion USD, the loss of equity value is then around 160 million USD given one standard deviation increase of the NTR gap (approximately 14%).

17 spreads, we should try to rule out the effect driven by firms’ market betas. As a result, I also compute another set of AARPNTR using market as the benchmark. Regressions 2, 4, and 6 in Table1 confirm that the empirical findings are not sensitive to the choice of models when computing firms’ abnormal returns. Similarly, the coefficient estimates are both statistically and economically significant. Given a one standard deviation increase in the firm’s NTR gap, the coefficient of -0.205 in column 2 translates to a reduction of 5.09% , or roughly 158 million USD in market value. In addition to the baseline results using equal weights when aggregating the product-level NTR gaps, Table1 also demonstrates that the negative and significant relations hold with respect to both value weighting methods using total imports, as shown in columns 3 and 4, and using the value of import from China, as shown in columns 5 and 6. For the total- import-value weighted specification, one standard deviation increase in firm-level PNTR tariff spread34 renders -0.164 and -0.182 standard deviation change in terms of CAPM and market-adjusted AAR, which translates to 4.55% (141 million USD) and 4.52% (140 million USD) loss in market value. Lastly for the China-import-value weighted version, the -0.140 and -0.159 standard deviation map into 3.89% (121 million USD) and 3.95% (122 million USD) CAPM and market-adjusted loss in shareholder’s value. Averaging across calculations over the six columns of Table1, we get a reduction of approximately 4.5% of the total market value, or 140 million USD over the 25 days in the event window. The regression results documented in Table1 are robust to adding firm-level controls. I further discuss this in the robustness section. Summing up the empirical findings from the U.S. equity market response, we see that firms more exposed to the PNTR tariff spread shock experience more negative return during the legislative event windows. This is consistent with the view that equity markets anticipating an increase in import competition following the PNTR bill will hurt U.S. firms that are more exposed to the shock—i.e., those with larger NTR gaps.

4.3.2 Chinese Market Response

As U.S. firms more exposed to the PNTR tariff shock experience larger reductions in market value, one a priori prediction is that their Chinese counterparts—firms also more exposed to the PNTR tariff shock—would relatively outperform in the stock market during the event dates. More PNTR-exposed U.S. firms take a hit, because they face more severe

34Both standard deviations are around 14% for the two diﬀerent value-weighted aggregation.

18 import competition from Chinese exporters into U.S.’s domestic market. In contrast, for Chinese firms with larger exposure to the PNTR tariff shock would gain, relative to firms with smaller PNTR tariff exposure, from disproportionately expanding into the U.S market. Surprisingly, Table2 documents the opposite results against our a priori reasoning. Sim- ilar to the findings in the U.S. market, there exists a negative and statistically significant relationship between Chinese firms’ average abnormal return and their NTR gaps, as vi- sualized by Figure4. This effect is present in all specifications, robust to either CAPM or market-adjusted abnormal return calculation, and persists under three versions of NTR weighting schemes.

[Figure 4 about here.]

[Table 2 about here.]

The economic magnitude in the Chinese market is smaller than that observed in the U.S. market. The coefficient estimate in column 1 of Table2, for example, means that one standard deviation increase in a firm’s NTR gap corresponds to a 0.085 standard deviation decrease in daily average abnormal return during the event windows. This translates to a total of 2.37% decline in market value, or about 93 million CNY (or 11 million USD) loss for shareholders.35 This is roughly 7%36 of the overall market value reduction (given a one standard deviation increase in the NTR gap) observed in the U.S. market. A mirror negative effect on the market value for more PNTR exposed firms in the Chinese market is puzzling at first sight, despite a much smaller economic magnitude compared with the U.S. findings. It is reasonable to expect an attenuated effect for the Chinese stock market, because China’s equity market was not as developed as and much smaller in size37 than U.S at that time. But the negative sign was unexpected a priori, considering that the

35From the regression estimate with standardized variables, we can multiple the point estimate of -0.085 PNTR by the standard deviation of AARCAP M , 1.12%, then multiply by 25 (number days of the event window), resulting in a total reduction of 2.37%. Considering that the average market value of a firm in 2000 in my sample is 3.9 billion CNY (or 0.45 billion USD), the loss of equity value is then around 93 million CNY (or 11 million USD), given a one standard deviation increase in the NTR gap (approximately 14%). 36The calculation in Section 4.3.1 shows, in the U.S. market, that a one standard deviation increase in firm’s equally weighted NTR gap corresponds to 160 million USD loss implied by CAPM-adjusted abnormal return calculation during the event windows. Here, for the Chinese market, 93 million CNY (or 11 million USD) is approximately 7% of the value reduction observed in the U.S. market. 37The first Chinese stock exchange was created in 1990. Around 2000, there were approximately 1,000 listed firms in the exchange, and the average market capitalization was 1.5 billion RMB (180 million dollars) for tradable shares and 4 billion RMB (483 million dollars) for total shares.

19 relative losses observed for U.S firms (with larger NTR gaps) are more vulnerable to import competition from Chinese importers. Hence, Chinese firms with a higher degree of potential export penetration into the U.S. market should benefit more. Why do we observe similar market value reduction for more PNTR exposed Chinese firms? If the PNTR tariff shock allows Chinese firms to gain access to the U.S. market, and to “steal” more U.S. market shares in larger NTR-gap industries, we should have observed a relative market value increase for larger NTR-gap firms. To further examine this observation, I take a closer look at the Chinese public firms in my sample. There are around 1,000 listed companies, approximately 700 of which are linked to the Manufacturing Survey from the NBS. Matched firms are either state-owned firms or large corporations with sales above 5 million RMB, and are likely to already be exporters. The PNTR shock has two effects on these Chinese firms. First, the resolution of tariff uncertainty increases firms’ willingness to expand in the U.S. market, because firms are less likely to revoke irreversible investment under tariff increase. On the other hand, the PNTR bill enables many smaller firms to enter the U.S. market. The latter negative competition effect from new entrants to the U.S. market dominates the first positive effect of pure uncertainty resolution. As a result, we could observe a relative hit in market value for firms with larger NTR gaps. The competition effect from new entrants comes from two sources. First, passage of the PNTR bill permanently eliminates the potential tariff increase, given the failure to renew the NTR trade waivers by Congress. This uncertainty resolution decreases the fixed cost of entering the U.S. market. Another important margin comes from the institutional reform embedded in the PNTR bill. As a prerequisite for accession to the WTO, the granting of PNTR status requires that China to comply with a series of regulations, including abolishing various forms of trade protectionism. One salient feature of the Chinese international trade sector before joining the WTO is that exporting business are heavily regulated and involved by the central government. The Ministry of Commerce provided export license and quotas to selected firms, most of which are state-owned. The PNTR bill demands the license be provided more freely, and allows more firms to participate in exporting markets. As a result, a large amount of non-SOEs, who weren’t allowed to export prior the PNTR bill, would finally start exporting. Without data on which of those publicly listed Chinese firms were already exporters to the U.S. market, firms’ administrative status could serve as a proxy because SOEs are more likely to obtain export license than non-SOEs due to the special

20 political favoritism norm in China. I then rerun the analysis for SOEs and non-SOEs. Table3 shows that SOEs with larger NTR gaps experience more reduction in market value. This is consistent with the narrative that SOEs might suffer from a loss of “protection” in the exporting business, because the PNTR bill also requires that the exporting license practice be replaced. This effect is more salient for firms in industries with more potential entries, due to the tariff uncertainty resolution—i.e., for those with larger NTR gaps.

[Table 3 about here.]

For non-SOEs, Table4 demonstrates slightly positive coefficient estimates, though not significant, on PNTR exposure. Since the number of non-SOEs is limited, we probably don’t have enough power to detect statistical significance. Moreover, the administrative type is not a perfect proxy for exporting status. Exchange-listed firms are large corporations and more likely to be exporters to begin with. A subsample of those non-SOEs with preexist- ing exporting status could be hurt just like SOE exporters. Hence the positive effects are attenuated for the non-SOEs. Figure5 summarizes the regression results and illustrates the effect of NTR gap on abnormal return for different types of publicly listed firms.

[Table 4 about here.]

[Figure 5 about here.]

Table5 formally tests the differential responses of market value between SOEs and non- SOEs. The interaction terms’ coefficients are positive and significant across all specifications. The interpretation is that an increase in NTR gap translates to different market responses between firm types. Column 1, for example, shows that non-SOEs outperform SOEs by 0.22 standard deviation given a one standard deviation increase in NTR gap. Alternatively, for firms operating in one industry (A) that has a 14% larger NTR gap than another industry (B), the magnitude of underperformance by SOEs versus non-SOEs is 6.2% more—roughly 216 million CNY (or 26 million USD)—in industry A than in industry B.38

38 PNTR Multiply the point estimate of -0.222 by the standard deviation of AARCAP M , 1.12%, then multiply by 25 (number days of the event window), resulting in a total reduction of 6.2%. Considering that the average market value of a non-SOE in 2000 in my sample is 3.5 billion CNY (or 0.42 billion USD), the loss of equity value is then around 216 million CNY (or 26 million USD) given a one standard deviation increase of the NTR gap (approximately 14%). If we average across all speciﬁcations in Table5, the magnitude reduces to approximately 3.6%, or 124 million CNY (or 15 million USD).

21 [Table 5 about here.]

I have shown that for firms operating in a larger NTR-gap industry, SOEs relatively underperform while non-SOEs overperform. This asymmetric response partially comes from the the relaxation of the export licensing practice, which mostly favored the SOEs prior to the PNTR bill. One caveat when interpreting the differential responses of SOEs and non-SOEs is that they differ according to their NTR gaps. Thus another, more important, factor that generates this additional layer of difference along the NTR gap dimension is that SOEs are less sensitive to the trade policy uncertainty shock than their non-SOE counterparts to begin with. Firms operating in a large NTR-gap industry face higher levels of policy uncertainty shock in case of a failure to renew the NTR status. SOEs, however, are less sensitive to this potential tariff increase than non-SOEs, because SOEs have implicit government subsidies and policy guarantees once the export environment turns unfavorable. As a result, the safety net provided by the government reduces the gain from pure policy uncertainty brought by the PNTR bill for SOEs. The competition effect dominates and makes SOEs suffer more in large NTR-gap industries. I will formalize this idea later when discussing the mechanism in Section5.

4.4 Robustness

I perform the following robustness tests. First, I investigate each legislative event window in addition to pooling the five windows and computing the average. The estimates for each window are consistent with prior beliefs about the voting outcomes. Details are discussed in the Appendices. Second, I control for firm attributes, and the negative relationship between firms’ AAR and their NTR gap remains for both U.S. and Chinese firms. Third, I experiment with a coarser definition of NTR exposure by looking at the major industry classification of firms. In other words, I ignore the product-level sales segment data when computing the NTR gap for each firm, and the main results barely change. Moreover, I investigate industry-level responses to the PNTR shock. Regressing market capitalization- weighted AAR of all firms operating in the same industry on the NTR gap still results in negative and significant relationships. A more comprehensive description of the robustness test, corresponding regression results, and discussions are included in the Appendices.

22 5 Mechanism: Theory

I have documented equity market responses to the PNTR bill for both U.S. and Chinese firms. In both markets, we observe that firms with larger NTR gaps experience larger reductions in firm value. The U.S. stock market reaction is more or less expected, because the PNTR bill grants more access to Chinese competitors and drives down profit for firms operating in industries with larger tariff-hike elimination. However, I observe similar stock market reactions for the Chinese firms, most of which are SOEs. To sum up, there are two unexpected findings from the Chinese side. First, SOEs don’t seem to respond in the expected direction, instead exhibiting larger decreases in firm value in larger NTR-gap industries, given the PNTR shock. Second, non-SOEs show insignificant, though positive, responses in the cross-section of NTR exposure. The surprising findings for the Chinese market can be rationalized by the following mechanism. SOEs and non-SOEs differ in their capacity to withstand negative shocks. The government tends to provide safety net protection for SOEs, especially in developing economies in which institutions are often inefficient. For example, Faccio, Masulis and McConnell(2006) document that the government undertakes state-contingent interventions that support politically connected firms. Hence, such favorable treatments received by SOEs make them less sensitive to potential tariff hikes before the PNTR bill than non-SOEs. In other words, if there were a sudden tariff increase once the annual NTR status was revoked by the U.S. Congress, Chinese SOEs would not bear as much of the negative shock as non-SOEs. The PNTR shock has two competing effects that affect Chinese firms’ entry into and profit from exports to the U.S. market. First, the PNTR bill phases out potential tariff increases and directly reduces the trade costs for Chinese firms that participate in the exporting business into the U.S.. Second, the elimination of tariff threats encourages more firms’ entry, and thus indirectly drives down firm profit though more intense competition; this is manifested through changing the price index I will model in Section 5.1. For SOEs, the first direct effect of trade cost reduction is relatively small, due to their insensitivity to the tariff- hike threats before the PNTR shock, as a result of government safety net provision, and the second indirect competition effect dominates. For non-SOEs, on the other hand, the first direct trade cost reduction overshadows the second indirect effect of competition. Therefore, more non-SOEs enter the exporting market, especially in large NTR-gap industries, and SOEs lose to the competition from more non-SOEs’ entry. In a nutshell, Figure6 panel (a)

23 shows the two competing direct and indirect eﬀects of the PNTR shock on both SOEs and non-SOEs, and panel (b) highlights the overall eﬀects before and after the PNTR shock. I will formalize the narrative in a simple qualitative model in Section 5.1.

5.1 Model

The setting features a Melitz(2003) type of heterogeneous firms. I describe the basic framework and consider two types of firms. One absorbs a partial impact of tariff increase, thanks to government safety net provisions, and the other takes in the full impact of the tariff-increase shock. We solve for both types of firms’ entry decisions and calculate their profit. A qualitative model with differential tariff shock absorption treatment between the two types of firms can generate predictions consistent with empirical findings for both in the equity market and the export market.

5.1.1 Demand Side

Consumers in the U.S. spend a ﬁxed share of income to consume one homogeneous good and spend the rest on a bucket of diﬀerentiated goods aggregated with constant elasticity of

substitution (CES). Consumers choose the optimal quantity of each diﬀerentiated good, qv, to maximize utility subject to their budget constraint. The resulting CES aggregate demand is σ−1 −σ qv = EP pv ,

where σ > 1 is the constant elasticity of substitution across varieties v, and pv is the consumer price. The aggregate demand shifter, E, is the total expenditure in the diﬀerentiated sector. 1 hR 1−σi 1−σ P = V ∈Ω (pv) is the CES price index for the set of available varieties, Ω.

5.1.2 Supply Side

Following the standard approach in the literature, there is a single factor (labor) with constant marginal productivity normalized to unity in the homogeneous good, which is taken as the numéraire such that the equilibrium wage is unity, w = 1. In the differentiated sector, there is a continuum of firms, each producing a variety v, and they compete monopolistically. Chinese firms pay fixed costs to export to the U.S. market. There are two types of Chinese

ﬁrms, state-owned and non-state-owned. The former pays sunk entry cost fs and the latter

24 sunk entry cost fn. The firms pick a draw of productivity z from a distribution with probability density function φ(z) 39; they know their underlying productivity and the distribution of other firms. In a similar fashion, to differentiate firm types, a state-owned firm draws productivity zs from distribution φ(zs), while a non-state-owned firm draws productivity zn from distribution φ(zn). The masses of state-owned and non-state-owned firm are ms and mn, respectively. Goods produced by Chinese firms are subject to an ad valorem tariff, τ ≥ 1 40, when being exported into the U.S. market. Before the PNTR bill, the prevailing tariff rate τ remains unchanged with probability π if the temporary NTR status is renewed by Congress. But with probability 1 − π, the NTR status is revoked and goods imported from China would face a higher tariff τh > τ. Nevertheless, the two types of Chinese exporters bear different effective tariff increases once the NTR status waiver is not approved. Due to the existence of various safety nets, only a fraction ψ of the increase in tariff will be borne by SOEs, while non-SOEs absorb the tariff increase all by themselves. After the PNTR bill, the probability of potential tariff hikes becomes zero (1 − π = 0).

5.1.3 Firm Entry

Chinese exporters observe the tariﬀ rate and choose p to maximize operating proﬁts, taking aggregate conditions as given, and correctly anticipating their equilibrium value. First-order conditions yield the equilibrium pricing rule with constant markup over cost, σ τw and thus the export price is σ−1 z . Substituting into the price aggregater, we derive the price index, which consists of three parts as shown in Equation4.

Z ∞ 1−σ Z ∞ 1−σ 1−σ 1−σ σ τw σ τw P = [P + ms φs(x)dx + mn φn(x)dx] (4) 0 ∗ ∗ zs σ − 1 x zn σ − 1 x

The price index combines components from U.S. domestic firms P0 and the contribution from both Chinese SOEs and non-SOEs. Firms are integrated over those that enter the ∗ ∗ export market above the productivity cutoffs (zs , zn). Chinese SOEs and non-SOEs make export decisions based on their productivity cutoff. ∗ ∗ There exists unique (zs , zn), such that SOEs (or non-SOEs) with productivity grater than ∗ ∗ zs (or zn) make positive profit and export. The cutoff choices are calculated such that the 39Note that since all firms with a given level of z behave identically, from now on I will index firms by productivity z instead of variety v. 40This is often called an icerberg transport cost.

25 ﬁrms make zero proﬁt:

0σ−1 σ w 1−σ 1 σ−1 σ w 1−σ 1 (1−π)EP ( (τ +(τh −τ))·ψ) +πEP ( τ) −fs = 0 (5) σ − 1 zs σ − 1 σ − 1 zs σ − 1

0σ−1 σ w 1−σ 1 σ−1 σ w 1−σ 1 (1−π)EP ( (τ +(τh−τ))·1) +πEP ( τ) −fn = 0 (6) σ − 1 zn σ − 1 σ − 1 zn σ − 1

Equations5 and6 correspond to SOEs and non-SOEs, respectively. In both equations5 and6, P denote the price index under the lower tariff regime and P 0 denote the price index under the higher tariff regime. There are two opposing effects that affect firms’ entry decision and profit. Figure6 illustrates the different responses for SOEs and non-SOEs under both direct and indirect effect of the PNTR.

[Figure 6 about here.]

5.2 Simulation Results: A Numerical Example

When there is no safety net provision for the SOEs, the tariff increase is borne 100% by the SOEs (ψ =1); The first row of Figure7, panel (a) shows that the growth in the number of SOE exporters and non-SOE exporters increases with respect to the NTR gap in a similar fashion. When most of the tariff-increase threat is not absorbed by the government, the growth in the number of exporters behaves differently. The first row of Figure7, panel (b) shows that more SOEs exit the export market in larger NTR-gap industries, but more non- SOEs enter in larger NTR-gap industries. The export value growths exhibit similar patterns, as shown in the second row of Figure7.

[Figure 7 about here.]

In addition the different export activity observed by the two types of firms, firms’ operating profits across different NTR-gap industries also behave differently. When we shut down the safety net channel, firms’ profits as a function of potential tariff hikes exhibit similar patterns, as shown in the third row of Figure7, panel (a). With safety net provisions, however, the third row of Figure7, panel (b) illustrates that a SOE’s profit decreases with the NTR tariff gap, and a non-SOE’s profit increases with the NTR gap.

26 5.3 Implications: Distributional Gains from Reallocation

Developing countries tend to have weak institutions. Robinson, Acemoglu and Johnson (2005) and LaPorta, Lopez-de Silanes and Shleifer(2008) provide abundant evidence on the relationship between institutions and long-term growth. SOEs are often considered a major source of corruption and inefficiency41, and a vast literature documents the preferential treatment of SOEs despite their lower productivity compared with non-SOEs (Khandelwal, Schott and Wei(2013)). In the context of U.S. granting PNTR status to China, I observe that industries a priori highly exposed to trade policy uncertainty shocks experience more non-SOEs’ entry into the export compared to SOEs. As a result, this asymmetric response with respect to the policy uncertainty shock generates additional distributional gains by alleviating misallocation distortions by inefficient institutions. The fact that SOEs receive preferential treatment from the government helps explain the mechanism behind how policy uncertainty elimination could foster welfare gains in the aggregate. In case of a negative shock—a sudden increase in tariffs for example—SOEs are less affected than their non-SOEs counterparts because the government usually steps in and helps SOEs avoid downside risk. This safety net provision makes SOEs less sensitive to NTR tariff gaps to begin with. Tariff uncertainty resolution thus has limited benefit for SOEs, while largely benefiting non-SOEs. Disproportionately more non-SOEs enter the market, which drives out inefficient SOEs, and hence improve aggregate productivity.

5.4 Alternative Mechanisms

I have so far proposed the mechanism of dampened tariﬀ elasticity due to safety net provisions for SOEs to rationalize the unexpected negative responses to the PNTR shock. One might think of other alternative explanations. In the following sections, I will examine these and rule out their legitimacy.

5.4.1 China’s Accession to the WTO

Congress granting PNTR to China paved the way for China’s accession to the WTO, since subjecting one WTO member to annual review of its NTR status—whether tied to

41Commentary from the Wall Street Journal: https://www.wsj.com/articles/ownership-is-key-to-fixing-chinas-soes-1466441214

27 specific emigration-related conditions or not—is clearly discriminatory. One may argue that China’s accession to the WTO might be a confounding factor in observing the empirical facts for the Chinese equity market. However, a closer look at the empirical strategy helps rule out the WTO story. First, China officially joined the WTO on 11 December 2001, while the legislative events around the China Trade bill were in 2000. Second, the main results from the stock return event study are in the cross-section, in essence. Firms with larger NTR gaps, in the cross-section, experienced lower AAR during the PNTR legislative events in 2000. In other words, the magnitude of underperformace in stock returns were more significant for firms with large NTR gaps. The NTR gap variation in the cross-section, originated from the Smoot-Hawley Act of 1930, is most likely to be exogenous to the China’s accession to the WTO in 2001. Moreover, it seems hard to imagine why firms’ NTR gaps would be correlated with unobserved factors affected by China’s joining the WTO, which would affect the AAR during the five PNTR legislative dates.

5.4.2 Endogenous Protection with Larger NTR Gaps

One may argue that U.S. firms operating in larger NTR-gap industries are protected against foreign competition to begin with. This might bias the estimation for both the U.S. and Chinese firms. Nonetheless, the higher tariffs were set in the 1930s, and most of the variations in NTR gaps come from this higher set of tariff rates. The long tariff legacy gives rise to a quasi-exogenous setting, and the sudden revocation of potential threat to this higher tariff regime is, arguably, a nice quasi-experiment that alleviates much of the heterogeneity concern.

5.4.3 Institutional Reform on Exporting Licenses

Around 2000, the Chinese government also implemented policy reform to relax export restrictions. Prior to 2000, SOEs had much less restrictive rules and were more likely to obtain export licenses than non-SOEs. After 2000, the export licensing rules became more homogeneous across different types of firms. One might suspect that the export license might play a large role in encouraging non-SOEs to enter the export business, and thus crowd out SOEs. This is certainly a plausible story. Still, it needs to go beyond the simple licensing story in order to explain why the negative effects for SOEs were more salient for firms with larger NTR gaps. The extent to which SOEs got hurt was related to their NTR gaps. If the

28 export license reform was common across industries, which seemed to be the case42, then license reform should not have differential effects across industries. One may further argue that in larger NTR-gap industries, the lifting of export license restrictions encourages more entry of non-SOEs, and thus put pressure on SOEs that drives down their profit. This story could work if we assume that all SOEs were already exporters before the PNTR bill43. One argument against this story is that we can do another event study and investigate whether firms’ AAR exhibit similar responses on the NTR gaps during licensing reform dates. Furthermore, the model I proposed in Section 5.1 helps me rule out this explanation. We may model the export license as the fixed cost of exporting for firms. Before the PNTR bill, SOEs and non-SOEs featured higher and lower fixed costs, and after the PNTR bill non-SOEs reduce their fixed costs to the level of SOEs. However, this is not enough to generate the empirical results observed in the equity market under an exhaustive combination of variable parameter choices. Intuitively, the direct benefits of reduced trade costs for SOEs are too large, such that the indirect competition effect doesn’t hurt them enough to generate loss in profits.

6 Mechanism: Empirical Tests

6.1 Evidence from Chinese Export Activities

I have documented that both U.S. and Chinese ﬁrms with larger NTR gaps experience relative decreases in market value. The Chinese market response could largely be driven by competition from new entrants in the exporting market, and more so in larger NTR-gap industries, which potentially attract more entry compared with lower NTR-gap industries. I examine the Chinese customs data and test whether more ﬁrm enter the U.S market in larger NTR-gap industries, which is also predicted by the model outlined in Section 5.1. Table6 presents evidence consistent with the model prediction. It shows that the number of non-SOE exporters increases more in high NTR-gap industries, while interestingly, the reverse is true for SOE exporters.

[Table 6 about here.]

42I don’t find any official documents that treat the licensing liberalization differently across industries; also, there was no mention of the NTR gaps. The details of the export license restrictions for SOEs and non-SOEs can be found in the Appendices. 43Or equivalently, the number of SOEs entering the export market is very small compared with non-SOEs after the PNTR bill because most SOEs are already exporters.

29 One point worth mentioning is that over the years, the total number of exporters increases significantly44 for both SOEs and non-SOEs. However, a simple comparison does not give us much economic insight, because there exist too many confounding factors that could affect the two types of firms differently—not to mention the overall trend whereby SOEs’ role is being downplayed by China’s increasing effort to integrate into the world economy. However, NTR gaps provide a relatively orthogonal margin that could separate the effect of competition triggered by trade liberalization, both in the form of tariff uncertainty resolution and other forms of trade restrictions. The PNTR bill fosters China’s institutional change in freeing up export restrictions, and the number of exporters would increase for both types of firms under a less restrictive regulatory regime. However, we observe from Table6 that this increase is proportional to industries’ NTR gaps—and, interestingly, in opposite directions for SOE and non-SOE exporters. In addition to observing the higher growth rate in the number of Chinese exporters in larger NTR-gap industries, we also find consistent evidence in the value of exports. I use a triple-difference setting, as shown in Equation7, to explore the total value of exports’ variation according to different NTR gaps (first difference) before and after the PNTR shock (second difference) and across the U.S. country dummy for the exporting destination (third difference) year 2000 to 2006.

ln(V )hct = β NTR Gaph × PostPNTRt × 1{c = U.S.}c + δct + δch + δht + α + εhct (7)

[Table 7 about here.]

The coefficient of interest in Equation7 is β. Country-year, country-product, and product- year fixed effects are included to control for potential unobserved endogenous variations that might contaminate the regression results. Table7 shows positive and statistically significant estimates of β for the whole sample of Chinese exporters. Moreover, this positive effect mainly comes from SOEs. For SOE exporters, the coefficient is close to zero, and we don’t observe a significant change in the value of exports on the interaction term. In other words, non-SOE exporters increase export value to the U.S. market more drastically in larger NTR- gap industries, while SOEs’ export values are not as sensitive to the NTR gaps to begin with.

44If we run an OLS regression without standardizing the variables, the intercept corresponds to the level change in the growth of exporters.

30 This is consistent with the conjecture that SOEs’ potential gains from tariﬀ uncertainty elimination is overshadowed by the competition spurred by new non-SOE entrants into the exporting market.

6.2 Trade Shock Elasticity under Safety Net Provision: Evidence from Equity Market Responses to Anti-dumping Cases

I have shown from the Chinese exporting activities that SOE exporters’ entry into the U.S. market is not sensitive to the NTR gap to begin with before the PNTR bill. I propose that this lack of sensitivity most likely comes from the preferential treatments for SOEs by the government. In particular, before the PNTR bill, a potential tariff increase threat is not salient for SOEs, because given a bad tariff shock, the government would step in and provide favorable policy aid; in contrast, for non-SOEs the tariff threat is fully borne without this safety net provision. In order to support the narrative of government safety net provisions under bad trade policy shocks, I further investigate whether SOEs experience smaller impact45 than non- SOEs in the face of bad trade shocks. Even though the potential tariff increase did not actually happen before the PNTR bill, there was credible threats of realization. As a result, an examination of how firms react under realized negative trade policy shock speaks directly to how firms would have behaved despite the threats of bad shocks never realized in terms of revoking the NTR status. The specific realized negative trade shocks I study here are anti-dumping investigations of Chinese imports into the U.S. market. U.S. policy makers initiate anti-dumping cases and impose hefty tariff afterward to seek protection for domestic products. For Chinese firms that export to the U.S. market, anti-dumping investigations shrink their business and hurt firms’ profits. However, SOEs are less affected than non-SOEs because the Chinese government could step in and provide favorable treatments or trading subsidies for example, and absorb the negative impact brought along by the anti-dumping investigations. I find consistent evidence in firms’ equity market performance. I collect anti-dumping investigation cases initiated by the U.S. against imported goods from China. I combine product-level firms’ export data from Chinese customs. I further

45This also provides empirical support for the variable ψ, dampened impact due to government’s safety net absorption of negative shocks, in the model presented in Section 5.1.

31 identify Chinese exporters that are publicly listed on Chinese stock exchanges. I then calculate the one-year lag share of export value affected by anti-dumping cases for each firm in the combined sample. I term this affected export share one year ahead of the anti-dumping investigation as “anti-dumping exposure.” Finally, I compute, for each firm, the average abnormal return (AAR) around each anti-dumping investigation case, and regress on a firm’s anti-dumping exposure. Equation8 is the specification of a panel regression of firms’ AAR on their anti-dumping exposure, including both firm fixed effects and time fixed effects to control for unobserved confounding variables that could potentially bias the estimates.

AARit = β0 Anti-dumping Exposurei +β1SOEi × Anti-dumping Exposurei +αi +αt +α+it (8)

[Table 8 about here.]

Table8 shows that when the SOE dummy is turned off, firms’ AAR is negative and significant under a higher anti-dumping exposure. This result is robust to varying choice of event window and not sensitive to various measures of anti-dumping calculation. However, the linear combination of the regression coefficients, which correspond to the anti-dumping exposure loading for SOEs, exhibit no significance. In other words, the anti-dumping investigation hurt Chinese exporters, but the effect is concentrated in non-SOEs and not significant for SOEs. This provides direct evidence to the narrative that SOEs do not absorb all of the negative impact of trade shocks, as their non-SOE counterparts do. Thus, SOEs are less sensitive to potential bad trade policy shocks to begin with.

7 Conclusion

This paper uses equity market responses to capture the net impact of firms’ exposure to a major U.S.-China trade liberalization event, the U.S. granting permanent normal trade relation (PNTR) to China in 2000. I analyze and compare U.S. and Chinese markets to infer winners and losers from trade, especially with respect to policy indeterminacy elimination shocks. U.S. firms operating in industries protected by higher potential tariff increases suffered greater market value losses once the tariff-hike threats were eliminated after the passage of the PNTR China Trade bill. Surprisingly, I find the same direction of response for Chinese

32 firms. In other words, Chinese firms facing higher potential tariff increases given a failure to renew favorable trading status by the U.S. incur larger amount of equity value reductions when the PNTR was granted. The Chinese market responses are unexpected at first sight, because PNTR granting by the U.S. removes the risk of potential tariff hikes, and thus helps Chinese firms expand into the U.S. market. Yet the equity market interprets the PNTR bill negatively. To rationalize this finding for the Chinese equity market, I show that the negative market responses concentrate mainly in SOEs. SOEs often receive favorable treatment from the government, and this safety net guarantee attenuates SOEs’ sensitivity to (potential) bad shocks, such as possible tariff hikes. As a result, the tariff rate uncertainty elimination benefits non-SOEs to a larger extent and encourages more firm entry into the export market. Hence, SOEs lose more due to increased competition from non-SOEs. I further provide evidence that non-SOEs exporting activities increase more in industries with larger potential tariff hikes, while this is not the case for SOEs. Given the well- documented evidence on the relative low productivity of SOEs versus non-SOEs, a larger amount of non-SOE exporters entering the market implies an overall gain in productivity. This paper also provides insight on how policy indeterminacy resolution can generate further distributional gains. Inefficient institutions tend to have misallocation of resources. Policy uncertainty elimination could then disproportionately benefit efficient but less protected firms. On the other hand, introducing more policy uncertainty could induce more rent for inefficient but protected firms. It will be an interesting extension for future research to investigate the implications of increased trade policy indeterminacy under the Trump administration.

33 References

Acemoglu, Daron, David Autor, David Dorn, Gordon H Hanson, and Brendan Price. 2016. “Import Competition and the Great US Employment Sag of the 2000s.” J. Labor Econ., 34(S1): S141–S198.

Auer, Raphael, and Andreas M Fischer. 2010. “The eﬀect of low-wage import competition on U.S. inﬂationary pressure.” J. Monet. Econ., 57(4): 491–503.

Autor, David H, David Dorn, and Gordon H Hanson. 2013. “The China Syndrome: Local Labor Market Eﬀects of Import Competition in the United States.” Am. Econ. Rev., 103(6): 2121–2168.

Autor, David H, David Dorn, and Gordon H Hanson. 2016. “The China Shock: Learning from Labor-Market Adjustment to Large Changes in Trade.” Annu. Rev. Econom., 8(1): 205–240.

Autor, David H, David Dorn, Gordon H Hanson, and Jae Song. 2014. “Trade Adjustment: Worker-Level Evidence.” Q. J. Econ., 129(4): 1799–1860.

Bansal, Ravi, Dana Kiku, Ivan Shaliastovich, and Amir Yaron. 2014. “Volatility, the Macroeconomy, and Asset Prices.” J. Finance, 69(6): 2471–2511.

Bloom, Nick. 2009. “The Impact of Uncertainty Shocks.” Econometrica, 77(3): 623–685.

Brandt, Loren, Johannes Van Biesebroeck, and Yifan Zhang. 2014. “Challenges of working with the Chinese NBS ﬁrm-level data.” China Econ. Rev., 30: 339–352.

Brandt, Loren, Johannes Van Biesebroeck, Luhang Wang, and Yifan Zhang. 2017. “WTO Accession and Performance of Chinese Manufacturing Firms.” Am. Econ. Rev., 107(9): 2784–2820.

Dixit, Avinash. 1989. “Entry and Exit Decisions under Uncertainty.” J. Polit. Econ., 97(3): 620–638.

Esposito, Federico, Marcelo Bianconi, and Marco Sammon. 2019. “Trade Policy Uncertainty and Stock Returns.”

34 Faccio, Mara, Ronald W Masulis, and John J McConnell. 2006. “Political Connec- tions and Corporate Bailouts.” J. Finance, 61(6): 2597–2635.

Feenstra, Robert C, John Romalis, and Peter K Schott. 2002. “U.S. Imports, Ex- ports, and Tariﬀ Data, 1989-2001.”

Greenland, Andrew, Mihai Ion, John Lopresti, and Peter K Schott. 2018. “Using Equity Market Reactions to Infer Exposure to Trade Liberalization.”

Griﬃn, Thomas P. 2018. “Globalization and U.S. Industry Concentration.”

Grossman, Gene M, and James A Levinsohn. 1989. “Import Competition and the Stock Market Return to Capital.” Am. Econ. Rev., 79(5): 1065–1087.

Hakobyan, Shushanik, and John McLaren. 2016. “Looking for Local Labor Market Eﬀects of NAFTA.” Rev. Econ. Stat., 98(4): 728–741.

Handley, Kyle, and Nuno Limão. 2015. “Trade and Investment under Policy Uncer- tainty: Theory and Firm Evidence.” American Economic Journal: Economic Policy, 7(4): 189–222.

Handley, Kyle, and Nuno Limão. 2017. “Policy Uncertainty, Trade, and Welfare: Theory and Evidence for China and the United States.” Am. Econ. Rev., 107(9): 2731–2783.

Hsieh, Chang-Tai, and Peter J Klenow. 2009. “Misallocation and Manufacturing TFP in China and India.” Q. J. Econ., 124(4): 1403–1448.

Huang, Yi, Chen Lin, Sibo Liu, and Heiwai Tang. 2019. “Trade Networks and Firm Value: Evidence from the US-China Trade War.”

Khandelwal, Amit K, Peter K Schott, and Shang-Jin Wei. 2013. “Trade Liberaliza- tion and Embedded Institutional Reform: Evidence from Chinese Exporters.” Am. Econ. Rev., 103(6): 2169–2195.

Krugman, Paul. 1980. “Scale Economies, Product Diﬀerentiation, and the Pattern of Trade.” Am. Econ. Rev., 70(5): 950–959.

Krugman, Paul. 1991. “Increasing Returns and Economic Geography.” J. Polit. Econ., 99(3): 483–499.

35 Krugman, Paul R. 1979. “Increasing returns, monopolistic competition, and international trade.” J. Int. Econ., 9(4): 469–479.

LaPorta, Rafael, Florencio Lopez-de Silanes, and Andrei Shleifer. 2008. “The Eco- nomic Consequences of Legal Origins.” J. Econ. Lit., 46(2).

Ma, Yue, Heiwai Tang, and Yifan Zhang. 2014. “Factor Intensity, product switching, and productivity: Evidence from Chinese exporters.” J. Int. Econ., 92(2): 349–362.

Melitz, Marc J. 2003. “The Impact of Trade on Intra-Industry Reallocations and Aggregate Industry Productivity.” Econometrica, 71(6): 1695–1725.

Pierce, Justin R, and Peter K Schott. 2009. “A Concordance Between Ten-Digit U.S. Harmonized System Codes and SIC/NAICS Product Classes and Industries.”

Pierce, Justin R, and Peter K Schott. 2016. “The Surprisingly Swift Decline of US Manufacturing Employment.” Am. Econ. Rev., 106(7): 1632–1662.

Robinson, James A, Daron Acemoglu, and Simon Johnson. 2005. “Institutions as a Fundamental Cause of Long-Run Growth.” Handbook of Economic Growth, 1A.

36 Figures

Across 6-digit NAICS Industries 3 2 Density 1 0 0 .2 .4 .6 .8 NTR rate - non-NTR rate

HS product aggregation by equal weight HS product aggregation by import value weight HS product aggregation by import (from China) value weight

Figure 1: Distribution of NTR Gaps (1999) Across NAICS Industries

This figure displays the distribution of normal trade relations (NTR) gap, the difference between NTR tariff rate and non-NTR tariff rate, across 6-digit (U.S.) NAICS industries in my sample. I obtain product-level NTR gap, classified using the Harmonized System (HS), from Feenstra, Romalis and Schott(2002) and aggregate to the NAICS industry level with three types of weighting schemes. Concordance from product HS code to the NAICS(1997 version) is provided by the U.S. Bureau of Economic Analysis (BEA) and Pierce and Schott(2009). HS product-level import from China to the U.S. is provided by UN Comtrade.

37 as Spoutlvlipr rmCiat h ..i rvddb NComtrade. UN NTR by level provided industry is the U.S. compute the Biesebroeck to to Van schemes China Brandt, weighting from provided and of import concordance (2009) types product-level the Schott three HS using employ and is gaps. industries I gap Pierce CIC NTR (2014). (BEA), Chinese Product-level Zhang Analysis to and non- Economic sample. then and of my and rate in Bureau NAICS tariff industries by NTR U.S. CIC between to (Chinese) difference mapped 4-digit the first across gap, NTR rate, of tariff distribution NTR the displays figure This

Density

iue2 itiuino T as(99 cosCCIndustries CIC Across (1999) Gaps NTR of Distribution 2: Figure 0 1 2 3 0 HS productaggregationbyimport(fromChina)valueweight HS productaggregationbyimportvalueweight HS productaggregationbyequalweight Across 4-digitCICIndustries .2 NTR rate-non-NTR 38 .4 .6 .8 Figure 3: U.S. Stock Market Response in the Cross Section

This figure is a binned scatter plot that illustrates the U.S. firms’ stock market performance during the event of U.S. granting permanent normal trade relations (PNTR) status to China. In the cross section, U.S. firms facing larger potential import competition from China, proxied by larger NTR gaps, experience lower average abnormal returns over all the legislative dates leading up to passage of the China Trade bill.

39 Figure 4: Chinese Stock Market Response in the Cross Section

This figure is a binned scatter plot that illustrates Chinese firms’ stock market performance during the event of U.S. granting permanent normal trade relations (PNTR) status to China. In the cross section, Chinese firms facing larger potential export penetration into the U.S. market, proxied by larger NTR gaps, experience lower average abnormal returns over all the legislative dates leading up to passage of the China Trade bill.

40 ae edn pt asg fteCiaTaebl.Terltosi spstv,tog not though positive, is relationship The bill. Trade export China potential the larger legislative firms. of the facing non-state-owned all passage Chinese firms over for to returns state-owned significant, abnormal firms’ Chinese up average Chinese lower section, leading experience of cross dates market regression U.S. the the the In into from penetration margins gaps. predictive NTR the their displays figure This iue5: Figure

(Average Abnormal Return) AAR (%) -.4 -.2 0 .2 .4 AAR 0 TR NT P (Degree ofPotentialExportPenetrationintheU.S.) Predictive Marginswith95%CIs aitosaogNRGpfrDffrn ye fFirms of Types Different for Gap NTR along Variations .2 .4 41 NTR Gap .6 State-Owned Firms Non-State-Owned Firms .8 AAR 1 TR NT P on (a) Direct Effect versus Indirect Effect (b) Before and After the PNTR Shock

Figure 6: Proﬁt, Productivity, and Entry for SOEs and Non-SOEs

This figure illustrates the overall effect of the China Trade bill on different types of Chinese firms. Panel (a) shows that SOEs experience smaller direct effect from tariff reduction than non-SOEs, while the indirect competition effect is similar for both SOEs and non-SOEs. The combined effect is shown in panel (b), where the cutoff productivity for entry increases for SOEs and decreases for non-SOEs. This translates to less entry of SOEs and more entry of non-SOEs. Moreover, given the same level of productivity, SOEs’ export profits decrease, while non-SOEs’ export profits increase.

42 (a) No Safety Net Provision (b) With Safety Net Provision

Figure 7: SOE Exporters versus Non-SOE Exporters

This ﬁgure compares Chinese SOE and non-SOE exporters for their growth in number, export value growth, and proﬁts. Panel (a) shows the comparison without safety net provision for SOEs. Panel (b) shows the comparison with safety net provision for SOEs.

43 Tables

Table 1: U.S. Firms’ Responses to the PNTR Shock: Baseline

(1) (2) (3) (4) (5) (6) PNTR PNTR PNTR PNTR PNTR PNTR AARCAP M AARMarket AARCAP M AARMarket AARCAP M AARMarket NTR Gap (ew) -0.186∗∗∗ -0.205∗∗∗ (0.052) (0.064)

NTR Gap (vw) -0.164∗∗∗ -0.182∗∗∗ (0.058) (0.068)

NTR Gap (vw CH) -0.140∗∗ -0.159∗∗ (0.060) (0.072) Observations 2,239 2,239 2,239 2,239 2,239 2,239 R2 0.035 0.042 0.027 0.033 0.020 0.025 This table contains estimates from firm-level regression of the overall daily average abnormal return (across all five legislative dates) on the NTR gap. Regressions 1, 3, and 5 use CAPM-adjusted average abnormal return (AAR), while regressions 2, 4, and 6 take market-adjusted AAR as the dependent variable. Firm-level NTR gap is calculated as the sales-weighted average NTR gap of all major segments’ 4-digit NAICS industry. The three rows in the regression table correspond to three different weighting schemes when aggregating product-level NTR gaps into industry-level gaps. Industry-level gaps are computed by equally weighting, total import value weighting or import value from China weighting the tariff spreads of all product lines classified under the same 4-digit NAICS industry. For diverse firms with product sales recorded in both good-producing and service industries, zero gap values are assigned to the service segments before constructing the sales-weighted average. The regression sample is restricted to firms with nonmissing return data during all five legislative dates. Regression variables are standardized to have zero mean and unit standard deviation. Robust standard errors, clustered at 4-digit NAICS level, are reported in parentheses. * p < 0.10, ** p < 0.05, ***p < 0.01

44 Table 2: Chinese Firms’ Responses to the PNTR Shock: Baseline

(1) (2) (3) (4) (5) (6) PNTR PNTR PNTR PNTR PNTR PNTR AARCAP M AARMarket AARCAP M AARMarket AARCAP M AARMarket NTR Gap (ew) -0.085∗∗ -0.076∗ (0.039) (0.042)

NTR Gap (vw) -0.081∗∗ -0.101∗∗∗ (0.034) (0.036)

NTR Gap (vw CH) -0.080∗∗ -0.111∗∗∗ (0.038) (0.034) Observations 714 714 714 714 714 714 R2 0.007 0.006 0.006 0.010 0.006 0.012 This table contains estimates from firm-level regression of the overall daily average abnormal return (across all five legislative dates) on the NTR gap. Regressions 1, 3, and 5 use CAPM-adjusted average abnormal return (AAR), while regressions 2, 4, and 6 take market-adjusted AAR as the dependent variable. Firms’ 4-digit Chinese Industry Classifications (CIC) are mapped into 6-digit NAICS industries, whose NTR gaps are computed by equally weighting, total import value weighting or import value from China weighting the tariff spreads of all product lines classified under the same 6-digit NAICS industry. The regression sample is restricted to firms with nonmissing return data during all five legislative dates. Regression variables are standardized to have zero mean and unit standard deviation. Robust standard errors, clustered at 4-digit CIC level, are reported in parentheses. * p < 0.10, ** p < 0.05, ***p < 0.01

45 Table 3: Chinese Firms’ Responses to the PNTR Shock: State-owned Firms

(1) (2) (3) (4) (5) (6) PNTR PNTR PNTR PNTR PNTR PNTR AARCAP M AARMarket AARCAP M AARMarket AARCAP M AARMarket NTR Gap (ew) -0.177∗∗∗ -0.118∗∗∗ (0.043) (0.042)

NTR Gap (vw) -0.129∗∗∗ -0.134∗∗∗ (0.034) (0.039)

NTR Gap (vw CH) -0.126∗∗∗ -0.134∗∗∗ (0.038) (0.039) Observations 546 546 546 546 546 546 This table contains estimates from firm-level regression of the overall daily average abnormal return (across all five legislative dates) on the NTR gap. Regressions 1, 3, and 5 use CAPM-adjusted average abnormal return (AAR), while regressions 2, 4, and 6 take market-adjusted AAR as the dependent variable. Firms’ 4-digit Chinese Industry Classifications (CIC) are mapped into 6-digit NAICS industries, whose NTR gaps are computed by equally weighting, total import value weighting or import value from China weighting the tariff spreads of all product lines classified under the same 6-digit NAICS industry. The regression sample is restricted to firms with nonmissing return data during all five legislative dates. Regression variables are standardized to have zero mean and unit standard deviation. Robust standard errors, clustered at 4-digit CIC level, are reported in parentheses. * p < 0.10, ** p < 0.05, ***p < 0.01

46 Table 4: Chinese Firms’ Response to PNTR Shock: Non-state-owned Firms

(1) (2) (3) (4) (5) (6) PNTR PNTR PNTR PNTR PNTR PNTR AARCAP M AARMarket AARCAP M AARMarket AARCAP M AARMarket NTR Gap (ew) 0.045 0.063 (0.035) (0.077)

NTR Gap (vw) 0.051 0.018 (0.037) (0.068)

NTR Gap (vw CH) 0.056 0.028 (0.042) (0.064) Observations 168 168 168 168 168 168 This table contains estimates from firm-level regression of the overall daily average abnormal return (across all five legislative dates) on the NTR gap. Regressions 1, 3, and 5 use CAPM-adjusted average abnormal return (AAR), while regressions 2, 4, and 6 take market-adjusted AAR as dependent variable. Firms’ 4-digit Chinese Industry Classifications (CIC) are mapped into 6-digit NAICS industries, whose NTR gaps are computed by equally weighting, total import value weighting or import value from China weighting the tariff spreads of all product lines classified under the same 6-digit NAICS industry. The regression sample is restricted to firms with nonmissing return data during all five legislative dates. Regression variables are standardized to have zero mean and unit standard deviation. Robust standard errors, clustered at 4-digit CIC level, are reported in parentheses. * p < 0.10, ** p < 0.05, ***p < 0.01

47 Table 5: Chinese Firms’ Responses to the PNTR Shock: Interaction

CAPM Market (1) (2) (3) (4) (5) (6) AARPNTR AARPNTR AARPNTR AARPNTR AARPNTR AARPNTR NTR Gap (ew) -0.177∗∗∗ -0.118∗∗∗ (0.043) (0.042) Non-State × Gap (ew) 0.222∗∗ 0.181∗∗ (0.092) (0.082) NTR Gap (vw) -0.129∗∗∗ -0.134∗∗∗ (0.034) (0.039) Non-State × Gap (vw) 0.180∗∗ 0.152∗∗ (0.085) (0.077) NTR Gap (vw CH) -0.126∗∗∗ -0.134∗∗∗ (0.038) (0.039) Non-State × Gap (vw CH) 0.182∗∗ 0.162∗∗ (0.082) (0.078) Observations 714 714 714 714 714 714 This table contains estimates from firm-level regression of the overall daily average abnormal return (across all five legislative dates) on the NTR gap. Regressions 1, 2, and 3 use CAPM-adjusted average abnormal return (AAR), while regressions 4, 5, and 6 take market-adjusted AAR as the dependent variable. Firms’ 4-digit Chinese Industry Classifications (CIC) are mapped into 6-digit NAICS industries, whose NTR gaps are computed by equally weighting, total import value weighting or import value from China weighting the tariff spreads of all product lines classified under the same 6-digit NAICS industry. The regression sample is restricted to firms with nonmissing return data during all five legislative dates. Regression variables are standardized to have zero mean and unit standard deviation. Robust standard errors, clustered at 4-digit CIC level, are reported in parentheses. * p < 0.10, ** p < 0.05, ***p < 0.01

48 Table 6: Number of Chinese Exporters within Each Industry

Growth in Number Growth in Number Non-SOE Exporters SOE Exporters NTR Gap (ew) 0.107∗∗ -0.123∗∗ (0.044) (0.057)

NTR Gap (vw) 0.120∗∗∗ -0.119∗∗ (0.043) (0.057)

NTR Gap (vw CH) 0.117∗∗ -0.118∗∗ (0.050) (0.056) Observations 486 486 486 489 489 489 This table contains estimates from industry-level regression of the growth in the number of exporters on the NTR gap. Regressions 1, 2, and 3 examine the growth rate of non-SOE exporters, while regressions 4, 5, 6 focus on those of SOE exporters. 6-digit NAICS industries’ NTR gaps are computed by equally weighting, total import value weighting or import value from China weighting the tariff spreads of all product lines classified under the same 6-digit NAICS industry. The exporter growth rates for each industry count the change in number of Chinese firms exporting to the U.S., whose exported product was classified under the 6-digit NAICS industry, over the average number of exporters in the beginning and the ending period of the sample (here from 2000 to 2006). The counts of exporting firms within each industry are adjusted for cases in which one 6-digit HS product corresponds to multiple 6-digit NAICS industries. Regression variables are standardized to have zero mean and unit standard deviation. Robust standard errors, clustered at 6-digit NAICS level, are reported in parentheses. * p < 0.10, ** p < 0.05, ***p < 0.01

49 Table 7: Product-Country-Year Export Value from China

ln(V) ln(V) ln(V) All Exporters SOE Exporters Non-SOE Exporters US × P ost × Gap 0.356∗∗∗ 0.027 0.291∗∗ (0.112) (0.118) (0.131) Fixed Effects (Product×Country) YES YES YES Fixed Effects (Product×Year) YES YES YES Fixed Effects (Year×Country) YES YES YES Observations 1,521,903 1,175,093 1,087,486 R2 0.787 0.720 0.703 This table contains estimates from product-country-year level triple difference-in-difference regression of log Chi- nese export value on the interaction of exporting destination being the United States indicator, post-PNTR dummy, and products’ 6-digit HS-level NTR gaps. All regression specifications are saturated and include a full sets of fixed effects, including product-country, product-year, and year-country fixed effects. Data span 2000 to 2006. Ro- bust standard errors, clustered at product-country level, are reported in parentheses. * p < 0.10, ** p < 0.05, ***p < 0.01

50 Table 8: Chinese Firms’ Response to Anti-dumping Investigations

(1) (2) (3) (4)

AAR[−2,2] AAR[0,5] AAR[0,5] AAR[0,5] AD Exposure (total export) -0.442∗ -0.581∗∗ (0.261) (0.286)

SOE × AD Exposure (total export) 0.297 0.375 (0.473) (0.456)

AD Exposure (U.S. export) -0.503∗ (0.267)

SOE × AD Exposure (U.S. export) 0.435 (0.283)

AD Exposure (Operating Revenue) -11.102∗∗ (5.341)

SOE × AD Exposure (Operating Revenue) 9.788 (9.991) Firm Fixed Effects Yes Yes Yes Yes Time Fixed Effects Yes Yes Yes Yes Linear Combination (AD Exposure for SOE) -0.14 -0.21 -0.07 -1.31 S.E. (0.39) (0.35) (0.10) (8.44) Observation 19,571 19,574 12,041 19,080 This table contains estimates from panel regression of firms’ daily average abnormal return around anti- dumping investigation initiation dates on their exposure to anti-dumping products in the previous year. The panel contains anti-dumping cases initiated from 2001 to 2013. Three different measures of anti-dumping exposure are used. First, the exported product value subject to anti-dumping investigations in the following year is divided by the total value of exported goods for a given firm within the same year. Second, the anti- dumping affected product export value is divided by the total value of exported goods to the U.S. market. Third, the export value is divided by the total operating revenue. All regression specifications include firm fixed effects and date fixed effects. Robust standard errors are reported in parentheses. * p < 0.10, ** p < 0.05, ***p < 0.01

51 Appendices

A Average Abnormal Return: U.S. Market

event PNTR To illustrate the computed AARi and AARi values across firms in the combined sample, Figures8 and9 compare the distribution of firms’ AAR by legislative events and by firm type, respectively. We may observe heterogeneous responses across firms and across events. Moreover, good-producing firms exhibit more left skewness than service firms.

[Figure 8 about here.]

[Figure 9 about here.]

I also compute the industry-level average abnormal return for each legislative event event PNTR AARj and the overall PNTR average abnormal return AARj . These industry-level average abnormal returns are calculated as the market capitalization-weighted average abnormal return of all companies whose largest segment is in industry j. Figure 10 compares industry- and ﬁrm-level PNTR average abnormal returns. It shows that within the same 6-digit NAICS industry in which ﬁrms operate, companies demonstrate heterogeneous responses to the PNTR shock.

[Figure 10 about here.]

B Average Abnormal Return: Chinese Market

event PNTR To illustrate the computed AARi and AARi values across firms in the combined sample, Figure 11 and 12 compare the distribution of firms’ AAR by legislative event and firm type, respectively. We may observe heterogeneous responses across firms and across events. Interestingly, compared with the U.S. counterpart pictured in Figure8 and9, we see a reversed skewness during the legislative events related to the House and Senate actions. Moreover, Chinese good-producing firms exhibit slightly more right skewness than service firms, while we observe more left skewness for U.S. good-producing firms.

[Figure 11 about here.]

[Figure 12 about here.]

52 Similar to the analysis of the U.S. market, I also compute the industry-level average ab- event normal return for each legislative event AARj and the overall PNTR average abnormal PNTR return AARj . These industry-level average abnormal returns are calculated as the market capitalization-weighted average abnormal returns of all companies whose largest segment is in industry j. Figure 13 compares industry- and ﬁrm-level PNTR average abnormal returns. It shows that within the same 4-digit CIC industry in which ﬁrms operate, companies demonstrate heterogeneous responses to the PNTR shock.

[Figure 13 about here.]

C Construction of Chinese Firms’ NTR Gaps

There are two ways to construct NTR gaps for Chinese firms. The first is to map each HS product to its corresponding U.S. NAICS industry using concordance provided by the Bureau of Economic Analysis (BEA) and Pierce and Schott(2009), and then convert to Chinese CIC industries using concordance by Brandt, Van Biesebroeck and Zhang(2014). After the concordance from HS product to CIC industries is completed, I compute weighted averages of NTR gaps for each CIC industry. The second method is to map each HS product directly to Chinese CIC industry using the concordance provided by Brandt et al.(2017). Similar to the first method, I then compute weighted averages of NTR gaps for each CIC industry. Figure 14 shows the distribution of NTR gaps under the two construction methods. Both methods produce similar NTR gap dispersion across CIC industries, with almost identical mean and standard deviation. In the main text, I use the first method because it is more relevant for a direct comparison between U.S. and Chinese industries through linked concordance. However, the second method is simpler in computation. My regression results are robust to both methods of NTR gap computation for Chinese firms.

[Figure 14 about here.]

D Industry-level NTR Gap

Table9 lists industries with the highest and lowest NTR gaps at the 6-digit NAICS level.

[Table 9 about here.]

53 E Robustness

E.1 Investigation of Each Legislative Event Window

Following Greenland et al.(2018), I also examine the relationship between firm-level NTR gap and average abnormal return for each of the five legislative events in addition to the average abnormal return across all five events. I only show the results using CAPM- adjusted average abnormal return and NTR gap calculated by equally weighting product’s tariff rates. The results are robust to alternative specifications, including employing market- adjusted abnormal return calculation and regression on value-weighting HS product, either by total import or by import from China. For the U.S firms, Table 10 shows negative and statistically significant relationships between a firm’s average abnormal returns and NTR gap during three of the five events. Similar to Greenland et al.(2018), introduction of the bill in the House of Representatives seems not to have a significant impact on firms’ stock market response. This is expected, since there are many legislative hurdles before a bill gets enacted into law. Since the introduction of the bill to the House is the first step and most bills are killed even before a vote46,I don’t expect to see a significant equity market response afterward. Then comes passage in the House, with 54% voting in favor. We observe a more negative and significant effect on the average abnormal return for firms with larger NTR gaps. Two months later, the cloture motion vote passed, 86 to 1247, which allowed the Senate to vote without further debate or filibuster. This event also records a negative and significant market response for NTR exposed firms. The following Senate vote, with 83% voting in favor, easily satisfied the simple majority rule. This result was expected, and I find less significant, though negative, stock responses48. Finally, the president signing the bill is the last step and we see similar market reactions. Averaging across all five events, column 6 documents a negative and significant relationship between a firms average abnormal return and its NTR gap, as documented in the Section 4.1.

[Table 10 about here.]

46A committee may vote to issue a report to the full chamber recommending that the bill be considered further. Only about one in four bills are reported out of committee. 47The Congressional Record can be found at https://www.congress.gov/congressional-record/2000/ 07/27/senate-section/article/S7768-2 48Greenland et al.(2018) also document a less signiﬁcant eﬀect during the senate vote window compared with the rest of the event dates.

54 For the Chinese firms, Table 11 shows similar results. All event windows demonstrate negative relationships, even though not all are statistically significant. The Chinese market event dates are one day behind the U.S market, and it is natural for Chinese investors not to be familiar with the legislative voting process in the U.S. Thus, the effect of the first three legislative hurdles are attenuated. When Congress officially passed the bill and it was later signed by President Clinton, there were larger and more significant market movements.

[Table 11 about here.]

E.2 Controlling for Firm Attributes

I examine the relationship between average abnormal return and PNTR exposure in the presence of firm attributes. Control variables that could potentially affect firms’ responses to the PNTR shock include commonly used proxies for firm’s investment and financing opportunities. Table 12 demonstrates that the negative and significant relation is robust to including various firm-level controls.

[Table 12 about here.]

E.3 Ignoring Product Segments

In the baseline regression, firm-level NTR gaps are computed as the sales-weighted average NTR gaps of the NAICS industries in which the product segments are categorized. For firms operating in multiple industries, this procedure produces a more accurate and detailed level of PNTR exposure for each firm. Now we take a step back and simply take the firm-level industry classification assigned by its major line of business without considering the exact breakdown into product-level segments. Table 13 shows that the baseline regression results still remain. The magnitude is slightly higher than that of the baseline, at which finer product-level segment sales are considered. Following the same calculation procedure49, we may average across the coefficient estimates of all specifications in Table 13. The calculated

49For example, in column 1, the coefficient of -0.208 means that a one standard deviation increase in firm’s NTR exposure (approximately 0.12 for equally weighted NTR gap without considering segment sales) translates to a 0.232% decrease in daily average abnormal return (-0.208 multiply one standard deviation of PNTR AARCAP M (1.116). Summing over 25 days in the event window, it becomes 5.8%, or 163 million USD (5.8% times the average market value of the firm, or 2.8 billion USD) loss in shareholder value. We may repeat this calculation for all six columns in Table 13.

55 average reduction in market value, given a one standard deviation increase in ﬁrm’s PNTR exposure, is around 5%, or 150 million USD.

[Table 13 about here.]

When we control for firm attributes, the negative and significant relationship between firm’s PNTR exposure and average abnormal return still holds, even if detailed product segment sales data are not used in computing the PNTR exposure. Similar to the specifications that consider a mix of multiple lines of business potentially in various industries, Table 14 shows that only with firms’ primary industry classification, regressions including firm attributes as controls also demonstrate robust findings.

[Table 14 about here.]

E.4 Industry-level Responses to the PNTR Shock

When we study the U.S market, each firm’s PNTR exposure is computed by sales-weight product segments’ industry NTR gap contribution. Section E.3 shows that the baseline results remain, even if a coarser firm-level industry classification is used without considering each product segment’s differential impact on firms’ exposure. This essentially credits the source of PNTR exposure variation to the industry level, to which each firm as a whole is assigned. I now further show the regression results taking industry as the unit of observation.

[Table 15 about here.]

Table 15 shows that even at the industry level, there exists a negative and significant relationship between exposure to PNTR and average abnormal return. The dependant variable, the average abnormal return of a 6-digit NAICS industry, is constructed as the market capitalization-weighted average abnormal return of all firms whose largest segment is within the 6-digit NAICS industry. In a similar fashion, I run the industry-level regression for the Chinese market. Table 16 shows that the results are very similar to firm-level regressions.

[Table 16 about here.]

56 F China’s Policy on Exporting Qualiﬁcations

Table 17 shows the export licensing policy in China from 1999 to 2000. We can see that state-owned firms face less restrictive exporting qualifications than non-state-owned firms.

[Table 17 about here.]

57 Figures

58 House .6 House Introduction House Vote Overall PNTR AAR

.4 Density

0 -7.5 -5 -2.5 0 2.5 5 7.5 Percent

Senate .6 Senate Cloture Senate Vote Overall PNTR AAR

.4 Density

0 -7.5 -5 -2.5 0 2.5 5 7.5 Percent

President .6 President Signing Overall PNTR AAR

.4 Density

0 -7.5 -5 -2.5 0 2.5 5 7.5 Percent

Figure 8: U.S. Stock Market Responses: PNTR Average Abnormal Returns by Legislative Event (%)

This ﬁgure presents the distributions of U.S. ﬁrms’ average abnormal returns across various legislative dates leading to passage of the China Trade bill, which grants permanent normal trade relations (PNTR) status to China. In order to improve readability, values below -7.5% and above 7.5% are dropped.

59 rd il hc rnspraetnra rd eain PT)sau oCia nodrto order In China. dropped. to are 7.5% status above (PNTR) and relations -7.5% trade below normal values firms’ readability, permanent U.S. improve grants of which service) and bill, Trade (good-producing types different of Type AAR distributions the Firm presents by figure Returns This Abnormal Average PNTR Responses: Market Stock (%) U.S. 9: Figure TR NT P vrg bomlrtrsars l eiltv ae edn opsaeo h China the of passage to leading dates legislative all across returns abnormal average , Density 0 .2 .4 .6 -7.5 -5 -2.5 60 Percent 0

2.5 5 Services Goods 7.5 -ii AC nuty nodrt mrv edblt,vle eo 75 n bv .%are 7.5% above and -7.5% below values readability, improve normal to permanent order grants In dropped. which industry. bill, Industry-level NAICS Trade 6-digit China China. the to of status capitalization-weighted passage (PNTR) to relations leading trade dates industry-level legislative and ﬁrm- all Ab- U.S. Average the compares PNTR ﬁgure Industry-level This versus Firm- Responses: (%) Market Returns Stock normal U.S. 10: Figure

Industry AAR -7.5 -5 -2.5 0 2.5 5 7.5 -7.5 -5 AAR -2.5 Firm AAR Goods TR NT P 0 2.5 fec r,woemjrslssgeti ihntespeciﬁc the within is segment sales major whose ﬁrm, each of 5 7.5 61

Industry AAR

AAR -7.5 -5 -2.5 0 2.5 5 7.5 -7.5 AAR TR NT P -5 TR NT P vrg bomlrtrs across returns, abnormal average , -2.5 Service Firm AAR scmue stemarket- the as computed is 0 2.5 5 7.5 eain PT)sau oCia nodrt mrv edblt,vle eo 75 n above and -7.5% below values trade readability, normal improve permanent various to grants order across which dropped. In returns bill, are Trade China. abnormal 7.5% China to average the status ﬁrms’ of (PNTR) passage Chinese relations to of leading distributions dates the legislative presents ﬁgure Legisla- This by Returns Abnormal Average PNTR (%) Responses: Event Market tive Stock Chinese 11: Figure

Density Density Density 0 .2 .4 .6 .8 1 0 .2 .4 .6 .8 1 0 .2 .4 .6 .8 1 -5 -5 -5 -2.5 -2.5 -2.5 President 62 Senate House Percent Percent Percent 0 0 0 2.5 2.5 2.5 Overall PNTRAAR Senate Vote Senate Cloture Overall PNTRAAR House Vote House Introduction Overall PNTRAAR President Signing 5 5 5 oipoeraaiiy ausblw-.%adaoe75 r dropped. order are In 7.5% above China. and to status -7.5% Chinese (PNTR) below relations of values trade readability, service) improve normal and permanent to grants (good-producing which bill types Trade China different of distributions Firm firms’ the by presents Returns figure Abnormal This Average PNTR Responses: Market (%) Stock Type Chinese 12: Figure AAR TR NT P Density 0 .2 .4 .6 vrg bomlrtrsars,allgsaiedtslaigt asg fthe of passage to leading dates legislative all across, returns abnormal average , -5 -2.5 63 Percent

0 2.5 5 Services Goods -ii I nuty nodrt mrv edblt,vle eo 75 n bv .%are 7.5% above and -7.5% below values readability, improve to normal order permanent In grants dropped. which industry. bill CIC Industry-level Trade 4-digit China China. the to of status capitalization-weighted passage (PNTR) to relations leading trade dates legislative industry-level Average and all ﬁrm- PNTR Chinese Industry-level compares ﬁgure versus This Firm- Responses: Market (%) Returns Stock Abnormal Chinese 13: Figure

Industry AAR -7.5 -5 -2.5 0 2.5 5 7.5 -7.5 -5 AAR -2.5 Firm AAR Goods TR NT P 0 fec r„woemjrln foeaini ihntespeciﬁc the within is operation of line major whose ﬁrm„ each of 2.5 5 7.5 64

Industry AAR

AAR -7.5 -5 -2.5 0 2.5 5 7.5 -7.5 AAR TR NT P -5 TR NT P vrg bomlrtrs across returns, abnormal average , -2.5 Service Firm AAR scmue stemarket- the as computed is 0 2.5 5 7.5 mot rmCiat h ..i rvddb NComtrade. I UN by (2017). provided product-level al. is HS U.S. et gaps. the Brandt NTR to by industry-level Chinese is China Brandt, the provided gap compute from to and concordance to NTR imports schemes then using (2009) product-level weighting of industry and the Schott types CIC method, three NAICS and employ mapping Chinese U.S. Pierce direct to BEA, to Under mapped and map- the (2014). mapped directly indirect rate Zhang by first Under and tariff provided is Biesebroeck NTR sample. concordance Van gap my between using NTR in difference industries industries product-level the CIC CIC the gaps, (Chinese) method, 4-digit NTR ping across of rate, distribution tariff the non-NTR displays figure This iue1:Dsrbto fNRGp cosCCIndustries CIC across Gaps NTR of Distribution 14: Figure

Density CIC Chinese to NAICS U.S. to HS mapping: Indirect (a) Density 0 1 2 3 0 1 2 3 0 0 b ietmpig St hns CIC Chinese to HS mapping: Direct (b) HS productaggregationbyimport(fromChina)valueweight HS productaggregationbyimportvalueweight HS productaggregationbyequalweight HS productaggregationbyimport(fromChina)valueweight HS productaggregationbyimportvalueweight HS productaggregationbyequalweight .2 Across 4-digitCICIndustries Across 4-digitCICIndustries .2 NTR rate-non-NTR NTR rate-non-NTR 65 .4 .4 .6 .6 .8 .8 Tables

66 Table 9: Industries with Highest and Lowest Gaps

(a) Industries with Highest Gaps Industry Naics Code NTR Gap Magnetic and Optical Recording Media Manufacturing 334613 0.80 Curtain and Drapery Mills 314121 0.79 Plastics Bottle Manufacturing 326160 0.77 Toilet Preparation Manufacturing 325620 0.75 Costume Jewelry and Novelty Manufacturing 339914 0.73

(b) Industries with Lowest Gaps Industry Naics Code NTR Gap Bituminous Coal Underground Mining 212112 0.00 Sugar Beet Farming 111991 0.00 Shellﬁsh Farming 112512 0.00 Wood Product Manufacturing 321114 0.00 Phosphatic Fertilizer Manufacturing 325312 0.00

67 Table 10: U.S. Firms’ Responses to the PNTR Shock: Each Legislative Event

(1) (2) (3) (4) (5) (6) AARHouseIntro AARHouseV ote AARSenateCloture AARSenateV ote AARClinton AARPNTR NTR Gap -0.028 -0.136∗∗∗ -0.118∗∗∗ -0.023 -0.170∗∗∗ -0.186∗∗∗ (0.041) (0.045) (0.026) (0.026) (0.043) (0.052) Observations 2,239 2,239 2,239 2,239 2,239 2,239 This table contains estimates from firm-level regression of daily average CAPM-adjusted abnormal return on the NTR gap. Regressions 1 to 5 correspond to five key legislative events, and regression 6 represents the average effect across five event dates. Firm-level NTR gap is calculated using the sales-weighted average NTR gap of all major segments’ 4-digit NAICS industry. Industry-level gaps are computed by equally weighting the tariff spreads of all product lines classified into the same 4-digit NAICS industry. For diverse firms with product sales recorded in both good-producing and service industries, zero gap values are assigned to the service segments before constructing the sales-weighted average. Firm sample is restricted to those with nonmissing return data during all of the five legislative dates. Regression variables are standardized to have zero mean and unit standard deviation. Robust standard errors, clustered at 4-digit NAICS level, are reported in parentheses. * p < 0.10, ** p < 0.05, *** p < 0.01

68 Table 11: Chinese Firms’ Responses to the PNTR Shock: Each Legislative Event

(1) (2) (3) (4) (5) (6) AARHouseIntro AARHouseV ote AARSenateCloture AARSenateV ote AARClinton AARPNTR NTR Gap -0.064 -0.038 -0.051 -0.080∗ -0.082∗∗ -0.085∗∗ (0.040) (0.035) (0.041) (0.041) (0.038) (0.039) Observations 714 714 714 714 714 714 This table contains estimates from firm-level regression of daily average CAPM-adjusted abnormal return on the NTR gap. Regressions 1 to 5 correspond to five key legislative events, and regression 6 represents the average effect across five event dates. Firm-level NTR gaps are assigned to match those of the 4-digit CIC industries, which are concorded from the 6-digit NAICS industries. Industry-level gaps are computed by equally weighting the tariff spreads of all product lines classified into the same 6-digit NAICS industry. Firm sample is restricted to those with nonmissing return data during all of the five legislative dates. Regression variables are standardized to have zero mean and unit standard deviation. Robust standard errors, clustered at 4-digit CIC level, are reported in parentheses. * p < 0.10, ** p < 0.05, *** p < 0.01

69 Table 12: U.S. Firms’ Responses to the PNTR Shock: With Firm Attributes

(1) (2) (3) (4) (5) (6) AARPNTR AARPNTR AARPNTR AARPNTR AARPNTR AARPNTR NTR Gap -0.205∗∗∗ -0.149∗∗ -0.135∗∗ -0.133∗∗ -0.133∗∗ -0.132∗∗ (0.064) (0.062) (0.060) (0.057) (0.057) (0.056)

Ln(PPE per Workier) 0.211∗∗∗ 0.305∗∗∗ 0.241∗∗∗ 0.199∗∗ 0.236∗∗∗ (0.047) (0.064) (0.066) (0.079) (0.090)

Ln(Mkt Cap) -0.121∗∗∗ -0.125∗∗∗ -0.093∗ -0.129∗∗ (0.045) (0.041) (0.049) (0.060)

CashF lows ∗∗∗ ∗∗∗ ∗∗∗ Assets 0.196 0.199 0.202 (0.070) (0.072) (0.073)

Book Leverage 0.051 0.051 (0.037) (0.038)

Tobins Q 0.045 (0.032) Observations 2,239 2,237 2,236 2,226 2,220 2,136 R2 0.042 0.084 0.090 0.125 0.127 0.130 This table contains estimates from firm-level regression of the overall daily market-adjusted average abnormal return (across all five legislative dates) on the NTR gap and firm attributes. Firm-level NTR gap is calculated as the sales-weighted average NTR gap of all major segments’ 4-digit NAICS industry. Industry-level gaps are computed by equally weighting the tariff spreads of all product lines classified into the same 4-digit NAICS industry. For diverse firms with product sales recorded in both good-producing and service industries, zero gap values are assigned to the service segments before constructing the sales-weighted average. The regression sample is restricted to firms with nonmissing return data during all of the five legislative dates. All accounting variables are winsorized at the 1% level. Regression variables are standardized to have zero mean and unit standard deviation. Robust standard errors, clustered at 4-digit NAICS level, are reported in parentheses. * p < 0.10, ** p < 0.05, *** p < 0.01

70 Table 13: U.S. Firms’ Responses to the PNTR Shock: Without Segments

(1) (2) (3) (4) (5) (6) PNTR PNTR PNTR PNTR PNTR PNTR AARCAP M AARMarket AARCAP M AARMarket AARCAP M AARMarket NTR Gap (ew) -0.208∗∗∗ -0.232∗∗∗ (0.055) (0.066)

NTR Gap (vw) -0.177∗∗∗ -0.197∗∗∗ (0.063) (0.074)

NTR Gap (vw CH) -0.145∗∗ -0.167∗∗ (0.067) (0.080) Observations 2,151 2,151 2,151 2,151 2,151 2,151 R2 0.043 0.054 0.031 0.039 0.021 0.028 This table contains estimates from firm-level regression of the overall daily average abnormal return (across all five legislative dates) on the NTR gap. Regressions 1, 3, and 5 use CAPM-adjusted average abnormal return (AAR), while regressions 2, 4, and 6 take market-adjusted AAR as the dependent variable. Firm-level NTR gap is equal to that of the 4-digit NAICS industry into which a firm is classified historically. The three rows in the regression table correspond to three different weighting schemes when aggregating product-level NTR gaps into industry-level gaps. Industry-level gaps are computed by equally weighting, total import value weighting, or import value from China weighting the tariff spreads of all product lines classified into the same 4-digit NAICS industry. The regression sample is restricted to firms with nonmissing return data during all five legislative dates. Regression variables are standardized to have zero mean and unit standard deviation. Robust standard errors, clustered at 4-digit NAICS level, are reported in parentheses. * p < 0.10, ** p < 0.05, *** p < 0.01

71 Table 14: U.S. Firms’ Responses to the PNTR Shock: With Firm Attributes, Ignoring Segments

(1) (2) (3) (4) (5) (6) AARPNTR AARPNTR AARPNTR AARPNTR AARPNTR AARPNTR NTR Gap -0.232∗∗∗ -0.165∗∗ -0.146∗∗ -0.143∗∗ -0.142∗∗ -0.143∗∗ (0.066) (0.067) (0.067) (0.064) (0.064) (0.063)

Ln(PPE per Workier) 0.201∗∗∗ 0.300∗∗∗ 0.237∗∗∗ 0.194∗∗ 0.233∗∗ (0.050) (0.067) (0.070) (0.084) (0.096)

Ln(Mkt Cap) -0.126∗∗∗ -0.128∗∗∗ -0.096∗ -0.136∗∗ (0.047) (0.044) (0.052) (0.064)

CashF lows ∗∗∗ ∗∗∗ ∗∗∗ Assets 0.187 0.191 0.194 (0.066) (0.069) (0.070)

Book Leverage 0.053 0.052 (0.040) (0.042)

Tobins Q 0.048 (0.032) Observations 2,151 2,150 2,149 2,139 2,134 2,056 R2 0.054 0.089 0.097 0.129 0.131 0.135 This table contains estimates from firm-level regression of the overall daily market-adjusted average abnormal return (across all five legislative dates) on the NTR gap and firm attributes. Firm-level NTR gap is equal to that of the 4-digit NAICS industry into which a firm is classified historically. Industry- level gaps are computed by equally weighting the tariff spreads of all product lines classified into the same 4-digit NAICS industry. The regression sample is restricted to firms with nonmissing return data during all five legislative dates. All accounting variables are winsorized at the 1% level. Regression variables are standardized to have zero mean and unit standard deviation. Robust standard errors, clustered at 4-digit NAICS level, are reported in parentheses. * p < 0.10, ** p < 0.05, *** p < 0.01

72 Table 15: U.S. Firms’ Responses to the PNTR Shock: Industry-level

(1) (2) (3) (4) (5) (6) PNTR PNTR PNTR PNTR PNTR PNTR AARCAP M AARMarket AARCAP M AARMarket AARCAP M AARMarket NTR Gap (ew) -0.119∗ -0.132∗∗ (0.065) (0.062)

NTR Gap (vw) -0.119∗ -0.116∗ (0.064) (0.061)

NTR Gap (vw CH) -0.144∗∗ -0.142∗∗ (0.064) (0.062) Observations 347 347 347 347 347 347 R2 0.014 0.017 0.014 0.013 0.021 0.020 This table contains estimates from industry-level regression of the overall daily average abnormal return (across all five legislative dates) on the NTR gap. Regressions 1, 3, and 5 use CAPM-adjusted average abnormal return (AAR), while regressions 2, 4, and 6 take market-adjusted AAR as the dependent variable. Industry-level abnormal return are the market capitalization-weighted average abnormal returns of all firms whose largest segment is within the same 6-digit NAICS industry. Industry-level NTR gaps are computed by equally weighting, total import value weighting, or import value from China weighting the tariff spreads of all product lines classified into the same 6-digit NAICS industry. The regression sample is restricted to firms with nonmissing return data during all five legislative dates. Regression variables are standardized to have zero mean and unit standard deviation. Robust standard errors, clustered at 6-digit NAICS level, are reported in parentheses. * p < 0.10, ** p < 0.05, *** p < 0.01

73 Table 16: Chinese Firms’ Responses to the PNTR Shock: Industry-level

(1) (2) (3) (4) (5) (6) PNTR PNTR PNTR PNTR PNTR PNTR AARCAP M AARMarket AARCAP M AARMarket AARCAP M AARMarket NTR Gap (ew) -0.133∗ -0.191∗∗ (0.075) (0.082)

NTR Gap (vw) -0.132∗∗ -0.126∗ (0.065) (0.075)

NTR Gap (vw CH) -0.120∗ -0.160∗∗ (0.067) (0.078) Observations 201 201 201 201 201 201 R2 0.018 0.036 0.011 0.016 0.014 0.025 This table contains estimates from industry-level regression of the overall daily average abnormal return (across all five legislative dates) on the NTR gap. Regressions 1, 3, and 5 use CAPM-adjusted average abnormal return (AAR), while regressions 2, 4, and 6 take market-adjusted AAR as the dependent variable. The industry-level abnormal returns are market capitalization-weighted average abnormal return of all the firms whose largest segment is within the 4-digit CIC industry. Industry-level NTR gaps are computed by equally weighting, total import value weighting, or import value from China weighting the tariff spreads of all product lines classified into the same 6-digit NAICS industry. Then the 6-digit NAICS industries are concorded to Chinese CIC industries. The regression sample is restricted to firms with nonmissing return data during all five legislative dates. Regression variables are standardized to have zero mean and unit standard deviation. Robust standard errors, clustered at 4-digit CIC level, are reported in parentheses. * p < 0.10, ** p < 0.05, *** p < 0.01

74 Table 17: Exporting Qualiﬁcations for SOEs and Non-SOEs in 1999–2000

State-owned Firms Non-state-owned Firms Minimal Registered Capital 5 Million CNY 8.5 Million CNY - special conditions (location) 3 Million CNY if MW NA - special conditions (institution) 2 Million CNY if RI NA - special conditions (product) 2 Million CNY if M&E NA

Additional Requirement None Required - minimal net assets NA 8.5 Million CNY - minimal sales NA 50 Milion CNY for 2 years - minimal export NA 1 Million USD

Approval Process Register Only Apply for Approval Source: China Ministry of Commerce